Undergraduate Texts in Mathematics Readings in Mathematics
Editors S. Axler F.W. Gehring K.A. Ribet Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002
This page intentionally left blank Gabor Toth
Glimpses of Algebra and Geometry
Second Edition
With 183 Illustrations, Including 18 in Full Color Gabor Toth Department of Mathematical Sciences Rutgers University Camden, NJ 08102 USA [email protected] Editorial Board S. Axler F.W. Gehring K.A. Ribet Mathematics Department Mathematics Department Mathematics Department San Francisco State East Hall University of California, University University of Michigan Berkeley San Francisco, CA 94132 Ann Arbor, MI 48109 Berkeley, CA 94720-3840 USA USA USA
Front cover illustration: The regular compound of five tetrahedra given by the face- planes of a colored icosahedron. The circumscribed dodecahedron is also shown. Computer graphic made by the author using Geomview. Back cover illustration: The regular compound of five cubes inscribed in a dodecahedron. Computer graphic made by the author using Mathematica.
Mathematics Subject Classification (2000): 15-01, 11-01, 51-01
Library of Congress Cataloging-in-Publication Data Toth, Gabor, Ph.D. Glimpses of algebra and geometry/Gabor Toth.—2nd ed. p. cm. — (Undergraduate texts in mathematics. Readings in mathematics.) Includes bibliographical references and index. ISBN 0-387-95345-0 (hardcover: alk. pape 1. Algebra. 2. Geometry. I. Title. II. Series. QA154.3 .T68 2002 512′.12—dc21 2001049269 Printed on acid-free paper.
2002, 1998 Springer-Verlag New York, Inc. All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Springer-Verlag New York, Inc., 175 Fifth Avenue, New York, NY 10010, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information stor- age and retrieval, electronic adaptation, computer software, or by similar or dissimi- lar methodology now known or hereafter developed is forbidden. The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights. Production managed by Francine McNeill; manufacturing supervised by Jeffrey Taub. Typeset from the author’s 2e files using Springer’s UTM style macro by The Bartlett Press, Inc., Marietta, GA. Printed and bound by Hamilton Printing Co., Rensselaer, NY. Printed in the United States of America. 987654321 ISBN 0-387-95345-0 SPIN 10848701 Springer-Verlag New York Berlin Heidelberg A member of BertelsmannSpringer Science+Business Media GmbH This book is dedicated to my students. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002
This page intentionally left blank Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002
...... Preface to the . . . . . Second Edition . . .
Since the publication of the Glimpses in 1998, I spent a consider- able amount of time collecting “mathematical pearls” suitable to add to the original text. As my collection grew, it became clear that a major revision in a second edition needed to be considered. In addition, many readers of the Glimpses suggested changes, clarifi- cations, and, above all, more examples and worked-out problems. This second edition, made possible by the ever-patient staff of Springer-Verlag New York, Inc., is the result of these efforts. Al- though the general plan of the book is unchanged, the abundance of topics rich in subtle connections between algebra and geometry compelled me to extend the text of the first edition considerably. Throughout the revision, I tried to do my best to avoid the inclusion of topics that involve very difficult ideas. The major changes in the second edition are as follows: 1. An in-depth treatment of root formulas solving quadratic, cubic, and quartic equations a` la van der Waerden has been given in a new section. This can be read independently or as preparation for the more advanced new material encountered toward the later parts of the text. In addition to the Bridge card symbols, the dagger † has been introduced to indicate more technical material than the average text. vii Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002
viii Preface to the Second Edition
2. As a natural continuation of the section on the Platonic solids, a detailed and complete classification of finite Mobius¨ groups ala` Klein has been given with the necessary background material, such as Cayley’s theorem and the Riemann–Hurwitz relation. 3. One of the most spectacular developments in algebra and geom- etry during the late nineteenth century was Felix Klein’s theory of the icosahedron and his solution of the irreducible quintic in terms of hypergeometric functions. A quick, direct, and modern approach of Klein’s main result, the so-called Normalformsatz, has been given in a single large section. This treatment is inde- pendent of the material in the rest of the book, and is suitable for enrichment and undergraduate/graduate research projects. All known approaches to the solution of the irreducible quin- tic are technical; I have chosen a geometric approach based on the construction of canonical quintic resolvents of the equation of the icosahedron, since it meshes well with the treatment of the Platonic solids given in the earlier part of the text. An al- gebraic approach based on the reduction of the equation of the icosahedron to the Brioschi quintic by Tschirnhaus transforma- tions is well documented in other textbooks. Another section on polynomial invariants of finite Mobius¨ groups, and two new appendices, containing preparatory material on the hyperge- ometric differential equation and Galois theory, facilitate the understanding of this advanced material. 4. The text has been upgraded in many places; for example, there is more material on the congruent number problem, the stereographic projection, the Weierstrass ℘-function, projective spaces, and isometries in space. 5. The new Web site at http://mathsgi01.rutgers.edu/∼gtoth/ Glimpses/ containing various text files (in PostScript and HTML formats) and over 70 pictures in full color (in gif format) has been created. 6. The historical background at many places of the text has been made more detailed (such as the ancient Greek approxima- tions of π), and the historical references have been made more precise. 7. An extended solutions manual has been created containing the solutions of 100 problems. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002
Preface to the Second Edition ix
I would like to thank the many readers who suggested improve- ments to the text of the first edition. These changes have all been incorporated into this second edition. I am especially indebted to Hillel Gauchman and Martin Karel, good friends and colleagues, who suggested many worthwhile changes. I would also like to ex- press my gratitude to Yukihiro Kanie for his careful reading of the text and for his excellent translation of the first edition of the Glimpses into Japanese, published in early 2000 by Springer- Verlag, Tokyo. I am also indebted to April De Vera, who upgraded the list of Web sites in the first edition. Finally, I would like to thank Ina Lindemann, Executive Editor, Mathematics, at Springer-Verlag New York, Inc., for her enthusiasm and encouragement through- out the entire project, and for her support for this early second edition.
Camden, New Jersey Gabor Toth Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002
This page intentionally left blank Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002
...... Preface to the . . . . . First Edition . . .
Glimpse: 1. a very brief passing look, sight or view. 2. a momentary or slight appearance. 3. a vague idea or inkling. —Random House College Dictionary
At the beginning of fall 1995, during a conversation with my re- spected friend and colleague Howard Jacobowitz in the Octagon Dining Room (Rutgers University, Camden Campus), the idea emerged of a “bridge course” that would facilitate the transition between undergraduate and graduate studies. It was clear that a course like this could not concentrate on a single topic, but should browse through a number of mathematical disciplines. The selection of topics for the Glimpses thus proved to be of utmost im- portance. At this level, the most prominent interplay is manifested in some easily explainable, but eventually subtle, connections be- tween number theory, classical geometries, and modern algebra. The rich, fascinating, and sometimes puzzling interactions of these mathematical disciplines are seldom contained in a medium-size undergraduate textbook. The Glimpses that follow make a humble effort to fill this gap.
xi Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002
xii Preface to the First Edition
The connections among the disciplines occur at various levels in the text. They are sometimes the main topics, such as Rational- ity and Elliptic Curves (Section 3), and are sometimes hidden in problems, such as the spherical geometric proof of diagonalization of Euclidean isometries (Problems 1 to 2, Section 16), or the proof of Euler’s theorem on convex polyhedra using linear algebra (Prob- lem 9, Section 20). Despite numerous opportunities throughout the text, the experienced reader will no doubt notice that analysis had to be left out or reduced to a minimum. In fact, a major source of difficulties in the intense 8-week period during which I pro- duced the first version of the text was the continuous cutting down of the size of sections and the shortening of arguments. Further- more, when one is comparing geometric and algebraic proofs, the geometric argument, though often more lengthy, is almost always more revealing and thereby preferable. To strive for some original- ity, I occasionally supplied proofs out of the ordinary, even at the “expense” of going into calculus a bit. To me, “bridge course” also meant trying to shed light on some of the links between the first recorded intellectual attempts to solve ancient problems of number theory, geometry, and twentieth-century mathematics. Ignoring detours and sidetracks, the careful reader will see the continuity of the lines of arguments, some of which have a time span of 3000 years. In keeping this continuity, I eventually decided not to break up the Glimpses into chapters as one usually does with a text of this size. The text is, nevertheless, broken up into subtexts corre- sponding to various levels of knowledge the reader possesses. I have chosen the card symbols ♣, ♦, ♥, ♠ of Bridge to indicate four levels that roughly correspond to the following: ♣ College Algebra; ♦ Calculus, Linear Algebra; ♥ Number Theory, Modern Algebra (elementary level), Geometry; ♠ Modern Algebra (advanced level), Topology, Complex Variables. Although much of ♥ and ♠ can be skipped at first reading, I encour- age the reader to challenge him/herself to venture occasionally into these territories. The book is intended for (1) students (♣ and ♦) who wish to learn that mathematics is more than a set of tools (the way sometimes calculus is taught), (2) students (♥ and ♠) who Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002
Preface to the First Edition xiii love mathematics, and (3) high-school teachers (⊂{♣, ♦, ♥, ♠}) who always had keen interest in mathematics but seldom time to pursue the technicalities. Reading what I have written so far, I realize that I have to make one point clear: Skipping and reducing the size of subtle arguments have the inherent danger of putting more weight on intuition at the expense of precision. I have spent a considerable amount of time polishing intuitive arguments to the extent that the more ex- perienced reader can make them withstand the ultimate test of mathematical rigor. Speaking (or rather writing) of danger, another haunted me for the duration of writing the text. One of my favorite authors, Iris Murdoch, writes about this in The Book and the Brotherhood,in which Gerard Hernshaw is badgered by his formidable scholar Lev- quist about whether he wanted to write mediocre books out of great ones for the rest of his life. (Tolearn what Gerard’s answer was, you need to read the novel.) Indeed, a number of textbooks iuenced me when writing the text. Here is a sample: 1. M. Artin, Algebra, Prentice-Hall, 1991; 2. A. Beardon, The Geometry of Discrete Groups, Springer-Verlag, 1983; 3. M. Berger, Geometry I–II, Springer-Verlag, 1980; 4. H.S.M. Coxeter, Introduction to Geometry, Wiley, 1969; 5. H.S.M. Coxeter, Regular Polytopes, Pitman, 1947; 6. D. Hilbert and S. Cohn-Vossen, Geometry and Imagination, Chelsea, 1952. 7. J. Milnor, Topology from the Differentiable Viewpoint, The Univer- sity Press of Virginia, 1990; 8. I. Niven, H. Zuckerman, and H. Montgomery, An Introduction to the Theory of Numbers, Wiley, 1991; 9. J. Silverman and J. Tate, Rational Points on Elliptic Curves, Springer-Verlag, 1992. Although I (unavoidably) use a number of by now classical ar- guments from these, originality was one of my primary aims. This book was never intended for comparison; my hope is that the Glimpses may trigger enough motivation to tackle these more advanced textbooks. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002
xiv Preface to the First Edition
Despite the intertwining nature of the text, the Glimpses contain enough material for a variety of courses. For example, a shorter ver- sion can be created by taking Sections 1 to 10 and Sections 17 and 19 to 23, with additional material from Sections 15 to 16 (treating Fuchsian groups and Riemann surfaces marginally via the exam- ples) when needed. A nonaxiomatic treatment of an undergraduate course on geometry is contained in Sections 5 to 7, Sections 9 to 13, and Section 17. The Glimpses contain a lot of computer graphics. The material can be taught in the traditional way using slides, or interactively in a computer lab or teaching facility equipped with a PC or a workstation connected to an LCD-panel. Alternatively, one can create a graphic library for the illustrations and make it accessi- ble to the students. Since I have no preference for any software packages (although some of them are better than others for par- ticular purposes), I used both Maple®1 and Mathematica®2 to create the illustrations. In a classroom setting, the link of either of these to Geomview3 is especially useful, since it allows one to manipu- late three-dimensional graphic objects. Section 17 is highly graphic, and I recommend showing the students a variety of slides or three- dimensional computer-generated images. Animated graphics can also be used, in particular, for the action of the stereographic pro- jection in Section 7, for the symmetry group of the pyramid and the prism in Section 17, and for the cutting-and-pasting technique in Sections 16 and 19. These Maple® text files are downloadable from my Web sites
http://carp.rutgers.edu/math-undergrad/science-vision.html and http://mathsgi01.rutgers.edu/∼gtoth/.
Alternatively, to obtain a copy, write an e-mail message to
1Maple is a registered trademark of Waterloo Maple, Inc. 2Mathematica is a registered trademark of Wolfram Research, Inc. 3A software package downloadable from the Web site: http://www.geom.umn.edu. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002
Preface to the First Edition xv or send a formatted disk to Gabor Toth, Department of Mathemat- ical Sciences, Rutgers University, Camden, NJ 08102, USA.
A great deal of information, interactive graphics, animations, etc., are available on the World Wide Web. I highly recommend scheduling at least one visit to a computer or workstation lab and explaining to the students how to use the Web. In fact, at the first implementation of the Glimpses at Rutgers, I noticed that my stu- dents started spending more and more time at various Web sites related to the text. For this reason, I have included a list of recom- mended Web sites and films at the end of some sections. Although hundreds of Web sites are created, upgraded, and terminated daily, every effort has been made to list the latest Web sites currently available through the Interent.
Camden, New Jersey Gabor Toth Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002
This page intentionally left blank Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002
...... Acknowledgments . . . . .
The second half of Section 20 on the four color theorem was writ- ten by Joseph Gerver, a colleague at Rutgers. I am greatly indebted to him for his contribution and for sharing his insight into graph theory. The first trial run of the Glimpses at Rutgers was during the first six weeks of summer 1996, with an equal number of under- graduate and graduate students in the audience. In fall 1996, I also taught undergraduate geometry from the Glimpses, covering Sec- tions 1 to 10 and Sections 17 and 19 to 23. As a result of the students’ dedicated work, the original manuscript has been revised and cor- rected, some of the arguments have been polished, and some extra topics have been added. It is my pleasure to thank all of them for their participation, enthusiasm, and hard work. I am particularly indebted to Jack Fistori, a mathematics education senior at Rutgers, who carefully revised the final version of the manuscript, making numerous worthwhile changes. I am also indebted to Susan Carter, a graduate student at Rutgers, who spent innumerable hours at the workstation to locate suitable Web sites related to the Glimpses. In summer 1996, I visited the Geometry Center at the University of Minnesota. I lectured about the Glimpses to an audience consisting of undergraduate and graduate students and high-school teachers. I wish to thank them for their valuable comments, which I took xvii Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002
xviii Acknowledgments
into account in the final version of the manuscript. I am especially indebted to Harvey Keynes, Education Director of the Geometry Center, for his enthusiastic support of the Glimpses. During my stay, I produced a 10-minute film Glimpses of the Five Platonic Solids with Stuart Levy, whose dedication to the project surpassed all my expectations. The typesetting of the manuscript started when I gave the first 20 pages to Betty Zubert as material with which a to practice LTEX. As the manuscript grew beyond any reasonable size, it is my pleasure to record my thanks to her for providing in- exhaustible energy that turned 300 pages of chicken scratch into a fine document.
Camden, New Jersey Gabor Toth Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002
...... Contents . . . . .
Preface to the Second Edition vii Preface to the First Edition xi Acknowledgments xvii Section 1 “A Number Is a Multitude Composed of Units”—Euclid 1 Problems 6 Web Sites 6 Section 2 “... There Are No Irrational Numbers at All”—Kronecker 7 Problems 21 Web Sites 25 Section 3 Rationality, Elliptic Curves, and Fermat’s Last Theorem 26 Problems 52 Web Sites 54 Section 4 Algebraic or Transcendental? 55 Problems 60 xix Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002
xx Contents
Section 5 Complex Arithmetic 62 Problems 71 Section 6 Quadratic, Cubic, and Quantic Equations 72 Problems 80 Section 7 Stereographic Projection 83 Problems 88 Web Site 89 Section 8 Proof of the Fundamental Theorem of Algebra 90 Problems 93 Web Site 95 Section 9 Symmetries of Regular Polygons 96 Problems 105 Web Sites 106 Section 10 Discrete Subgroups of Iso (R2)107 Problems 120 Web Sites 121 Section 11 Mobius¨ Geometry 122 Problems 130 Section 12 Complex Linear Fractional Transformations 131 Problems 137 Section 13 “Out of Nothing I Have Created a New Universe”—Bolyai 139 Problems 156 Section 14 Fuchsian Groups 158 Problems 171 Section 15 Riemann Surfaces 173 Problems 197 Web Site 198 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002
Contents xxi
Section 16 General Surfaces 199 Problems 208 Web Site 208 Section 17 The Five Platonic Solids 209 Problems 248 Web Sites 254 Film 254 Section 18 Finite Mobius¨ Groups 255
Section 19 Detour in Topology: Euler–Poincare´ Characteristic 266 Problems 278 Film 278 Section 20 Detour in Graph Theory: Euler, Hamilton, and the Four Color Theorem 279 Problems 294 Web Sites 297 Section 21 Dimension Leap 298 Problems 304 Section 22 Quaternions 305 Problems 315 Web Sites 316 Section 23 Back to R3!317 Problems 328 Section 24 Invariants 329 Problem 344 Section 25 The Icosahedron and the Unsolvable Quintic 345 A. Polyhedral Equations 346 B. Hypergeometric Functions 348 C. The Tschirnhaus Transformation 351 D. Quintic Resolvents of the Icosahedral Equation 355 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002
xxii Contents
E. Solvability of the Quintic a` la Klein 363 F. Geometry of the Canonical Equation: General Considerations 365 G. Geometry of the Canonical Equation: Explicit Formulas 369 Problems 377 Section 26 The Fourth Dimension 380 Problems 394 Film 395 Appendix A Sets 397 Appendix B Groups 399 Appendix C Topology 403 Appendix D Smooth Maps 407 Appendix E The Hypergeometric Differential Equation and the Schwarzian 409 Appendix F Galois Theory 419 Solutions for 100 Selected Problems 425 Index 443 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002
. . . . . “A Number Is a . . . . . Multitude Composed . . . 1 . SECTION . of Units”—Euclid
♣ We adopt Kronecker’s phrase: “God created the natural numbers, and all the rest is the work of man,” and start with the set N {1, 2, 3, 4, 5, 6,...} of all natural numbers. Since the sum of two natural numbers is again a natural number, N carries the operation1 of addition + : N × N → N.
Remark. Depicting natural numbers by arabic numerals is purely tradi- tional. Romans might prefer N {I, II, III, IV, V, VI,...}, and computers work with N {1, 10, 11, 100, 101, 110,...}. Notice that converting a notation into another is nothing but an iso- morphism between the respective systems. Isomorphism respects
1If needed, please review “Sets” and “Groups” in Appendices A and B. 1 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002
2 1. “A Number Is a Multitude Composed of Units”—Euclid
addition; for example, 29+33 62 is the same as XXIX +XXXIII LXII or 11101 + 100001 111110.
From the point of view of group theory, N is a failure; it does not have an identity element (that we would like to call zero) and no element has an inverse. We remedy this by extending N to the (additive) group of integers Z {0, ±1, ±2, ±3, ±4, ±5, ±6,...}. Z also carries the operation of multiplication × : Z × Z → Z. Since distributivity holds, Z forms a ring with respect to addition and multiplication. Although we have 1 as the identity element with respect to ×, we have no hope for Z to be a multiplicative group; remember the saying: “Thou shalt not divide by zero!” To remedy this, we delete the ominous zero and consider Z# Z −{0} {±1, ±2, ±3, ±4, ±5, ±6,...}. The requirement that integers have inverses gives rise to fractions or, more appropriately, rational numbers: Q Q # ∪{0} {a/b | a, b ∈ Z#}∪{0}, where we put the zero back to save the additive group structure. All that we learned in dealing with fractions can be rephrased ele- gantly by saying that Q is a field: Q is an additive group, Q # is an abelian (i.e., commutative) multiplicative group, and addition and multiplication are connected through distributivity. After having created Z and Q , the direction we take depends largely on what we wish to study. In elementary number theory, when studying divisibility properties of integers, we consider, for a given n ∈ N, the (additive) group Zn of integers modulo n. The simplest way to understand
Zn Z/nZ {[0], [1],...,[n − 1]} is to start with Z and to identify two integers a and b if they differ by a multiple of n. This identification is indicated by the square bracket; [a] means a plus all multiples of n. Clearly, no numbers are identified among 0, 1,...,n− 1, and any integer is identified Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002
1. “A Number Is a Multitude Composed of Units”—Euclid 3
0 n 2n Figure 1.1
with exactly one of these. The (additive) group structure is given by the usual addition in Z. More explicitly, [a] + [b] [a + b], a, b ∈ Z. Clearly, [0] is the zero element in Zn, and −[a] [−a] is the additive inverse of [a] ∈ Zn. Arithmetically, we use the division algorithm to find the quotient q and the remainder 0 ≤ r a qn + r, and set [a] [r]. The geometry behind this equality is clear. Con- sider the multiples of n, nZ ⊂ Z,asaone-dimensional lattice (i.e., an infinite string of equidistantly spaced points) in R as in Figure 1.1. Now locate a and its closest left neighbor qn in nZ (Figure 1.2). The distance between qn and a is r, the latter between 0 and n − 1. Since a and r are to be identified, the following geometric picture emerges for Zn:WrapZ around a circle infinitely many times so that the points that overlap with 0 are exactly the lattice points in nZ; this can be achieved easily by choosing the radius of the circle to be n/2π. Thus, Zn can be visualized as n equidistant points on the perimeter of a circle (Figure 1.3). Setting the center of the cir- cle at the origin of a coordinate system on the Cartesian plane R2 such that [0] is the intersection point of the circle and the posi- tive first axis, we see that addition in Zn corresponds to addition of angles of the corresponding vectors. A common convention is to choose the positive orientation as the way [0], [1], [2],...increase. This picture of Zn as the vertices of a regular n-sided polygon (with angular addition) will recur later on in several different contexts. (q − 1)naqn (q + 1)n Figure 1.2 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 4 1. “A Number Is a Multitude Composed of Units”—Euclid Figure 1.3 Remark. In case you’ve ever wondered why it was so hard to learn the clock 2 in childhood, consider Z60. Why the Babylonian choice of 60? Con- sider natural numbers between 1 and 100 that have the largest possible number of small divisors. ♥ The infinite Z and its finite offsprings Zn, n ∈ N, share the basic property that they are generated by a single element, a prop- erty that we express by saying that Z and Zn are cyclic. In case of Z, this element is 1 or −1; in case of Zn, a generator is [1]. ♣ You might be wondering whether it is a good idea to reconsider multiplication in Zn induced from that of Z. The answer is yes; multiplication in Z gives rise to a well-defined multiplication in Zn by setting [a] · [b] [ab], a, b ∈ Z. Clearly, [1] is the multiplicative # identity element. Consider now multiplication restricted to Zn Zn −{[0]}. There is a serious problem here. If n is composite, that # is, n ab, a, b ∈ N, a, b ≥ 2, then [a], [b] ∈ Zn, but [a] · [b] [0]! # Thus, multiplication restricted to Zn is not even an operation. 2Actually, a number system using 60 as a base was developed by the Sumerians about 500 years before it was passed on to the Babylonians around 2000 b.c. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 1. “A Number Is a Multitude Composed of Units”—Euclid 5 We now pin our hopes on Zp, where p a prime. Elementary num- ber theory says that if p divides ab, then p divides either a or b. # This directly translates into the fact that Zp is closed under multi- plication. Encouraged by this, we now go a step further and claim # that Zp is a multiplicative group! Since associativity follows from associativity of multiplication in Z, it remains to show that each el- # ement a ∈ Zp has a multiplicative inverse. To prove this, multiply the complete list [1], [2],...,[p − 1] by [a] to obtain [a], [2a],...,[(p − 1)a]. # By the above, these all belong to Zp. They are mutually disjoint. Indeed, assume that [ka] [la], k, l 1, 2,...,p− 1. We then have [(k − l)a] [k − l] · [a] [0], so that k l follows. Thus, the list # above gives p − 1 elements of Zp. But the latter consists of exactly p − 1 elements, so we got them all! In particular, [1] is somewhere in this list, say, [aa¯ ] [1], a¯ 1,...,p− 1. Hence, [a¯] is the mul- tiplicative inverse of [a]. Finally, since distributivity in Zp follows from distributivity in Z, we obtain that Zp is a field for p prime. We give two applications of these ideas: one for Z3 and another 2 2 for Z4. First, we claim that if 3 divides a + b , a, b ∈ Z, then 3 divides both a and b. Since divisibility means zero remainder, all we have to count is the sum of the remainders when a2 and b2 are divided by 3. In much the same way as we divided all integers to even (2k) and odd (2k + 1) numbers (k ∈ Z), we now write a 3k, 3k + 1, 3k + 2 accordingly. Squaring, we obtain a2 9k2, 9k2 + 6k + 1, 9k2 + 12k + 4. Divided by 3, these give remainders 0 or 1, with 0 corresponding to a being a multiple of 3. The situation is the same for b2. We see that when dividing a2 + b2 by 3, the possible remainders are 0 + 0, 0 + 1, 1 + 0, 1 + 1, and the first corresponds to a and b both being multiples of 3. The first claim follows. Second, we show the important number theoretical fact that no number of the form 4m + 3 is a sum of two squares of integers. (Notice that, for m 0, this follows from the first claim or by inspection.) This time we study the remainder when a2 + b2, a, b ∈ Z, is divided by 4. Setting a 4k, 4k + 1, 4k + 2, 4k + 3, a2 gives remainders 0 or 1. As before, the possible remainders for a2 + b2 are 0 + 0, 0 + 1, 1 + 0, 1 + 1. The second claim also follows. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 6 1. “A Number Is a Multitude Composed of Units”—Euclid ♥ The obvious common generalization of the two claims above is also true and is a standard fact in number theory. It asserts that if a prime p of the form 4m +3 divides a2 +b2, a, b ∈ Z, then p divides both a and b. Aside from the obvious decomposition 2 12 + 12, the primes that are left out from our considerations are of the form 4m + 1. A deeper result in number theory states that any prime of the form 4m + 1 is always representable as a sum of squares of two integers. Fermat, in a letter to Mersenne in 1640, claimed to have a proof of this result, which was first stated by Albert Girard in 1632. The first published verification, due to Euler, appeared in 1754. We postpone the proof of this result till the end of Section 5. Problems 1. Use the division algorithm to show that (a) the square of an integer is of the form 3a or 3a + 1, a ∈ Z; (b) the cube of an integer is of the form 7a,7a + 1 or 7a − 1, a ∈ Z. 2. Prove that if an integer is simultaneously a square and a cube, then it must be of the form 7a or 7a + 1. (Example: 82 43.) 3. (a) Show that [2] has no inverse in Z4. (b) Find all n ∈ N such that [2] has an inverse in Zn. 4. Write a ∈ N in decimal digits as a a1a2 ···an, a1,a2,...,an ∈{0, 1,...,9}, a1 0. Prove that [a] [a1 + a2 +···+an]inZ9. 5. Let p>3 be a prime, and write 1 + 1 +···+ 1 12 22 (p − 1)2 as a rational number a/b, where a, b are relatively prime. Show that p|a. Web Sites 1. www.utm.edu/research/primes 2. daisy.uwaterloo.ca/∼alopez-o/math-faq/node10.html Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 ...... “ There Are No . . . . . Irrational Numbers . . . 2 . SECTION . at All”—Kronecker ♣ Although frequently quoted, the epigraph to this chapter and some other statements of Kronecker on irrational numbers have been shown to be distortions.1 (See H.M. Edwards’s articles on Kronecker in History and Philosophy of Modern Mathematics (Min- neapolis, MN, 1985) 139–144, Minnesota Stud. Philos. Sci., XI, Univ. Minnesota Press, Minneapolis, MN, 1988.) In calculus, the field of rational numbers Q is insufficient for several reasons, including convergence. Thus, Q is extended to the field of real numbers R. It is visualized by passing the Cartesian Bridge2 connecting algebra and geometry; each real number (in infinite decimal representation) corresponds to a single point on the real line. Do we actually get new numbers? Here is a simple answer known to Pythagoras (c. 570–490 b.c.): 1I am indebted to Victor Pambuccian for pointing this out. 2Cartesius is the Latinized name of Rene´ Descartes, who, contrary to widespread belief, did not invent the coordinate axes, much less analytic geometry. Here we push this gossip a little further. Note that analytic geometry was born in Fermat’s Introduction to Plane and Solid Loci in 1629; although circulated from 1637 on, it was not published in Fermat’s lifetime. The notion of perpendicular coordinate axes can be traced back to Archimedes and Apollonius. Both Descartes and Fermat used coordinates but only with nonnegative values; the idea that coordinates can also take negative values is due to Newton. 7 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 8 2. “... There Are No Irrational Numbers at All”—Kronecker Proposition√ 1. 2(the unique positive number whose square is 2) is not rational. Proof. √ √ ¬3 Assume that 2 is rational; i.e., 2 a/b for some a, b ∈ Z.We may assume that a and b are relatively prime, since otherwise we cancel the common factors in a and b. Squaring, we get a2 2b2. A glimpse of the right-hand side shows that a2 is even. Thus a must be even, say, a 2c. Then a2 4c2 2b2. Hence b2 and b must be even. Thus 2 is a common factor of a and b. ¬ Remark. √ Replacing 2 by any prime p, we see that p is not rational. Instead of repeating the argument above, we now describe a more powerful result due to Gauss. (If you study the following proof carefully, you will see that it is a generalization of the Pythagorean argument above.) The idea√ is very simple (and will reoccur later) and is based p x2 −p on the fact that is a solution√ of the quadratic equation 0. To show irrationality of p, we study the rational solutions of this equation. More generally, assume that the polynomial equation n P(x) c0 + c1x +···+cnx 0,cn 0, with integer coefficients c0,c1,...,cn ∈ Z, has a rational root x a/b, a, b ∈ Z. As usual, we may assume that a and b are relatively prime. Substituting, we have n c0 + c1(a/b) +···+cn(a/b) 0. Multiplying through by bn−1, we obtain n−1 n−2 n−1 n c0b + c1ab +···+cn−1a + cna /b 0. n This says that cna /b must be an integer, or equivalently, b divides n cna . Since a and b are relatively prime, we conclude that b divides 3¬ indicates indirect argument; that is, we assume that the statement is false and get (eventually) a contradiction (indicated by another ¬). Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 2. “... There Are No Irrational Numbers at All”—Kronecker 9 n cn. In a similar vein, if we multiply through by b /a,weget n n−1 n−1 c0b /a + c1b +···+cna 0, and it follows that a divides c0. Specializing, we see that if x a/b is a solution of xn c, then b ±1 and so c (±a)n. Thus, if c is not the nth power of an integer, then xn c does not have any rational solutions. This is indeed a vast generalization of the Pythagorean argument above! Algebraically, we think of a real number as an infinite deci- mal. This decimal representation is unique (and thereby serves as a definition), assuming that we exclude representations termi- nating in a string of infinitely many 9’s; for example, instead of 1.2999 ...we write 1.3. How can we recognize the rational num- bers in this representation? Writing 1/3as0.333 ...gives a clue to the following: Proposition 2. An infinite decimal represents a rational number iff it terminates or repeats. Proof. We first need to evaluate the finite geometric sum 1 + x + x2 +···+xn. For x 1, this is n + 1, hence we may assume that x 1. This sum is telescopic; in fact, after multiplying through by 1 − x, everything cancels except the first and last term. We obtain (1 − x)(1 + x + x2 +···+xn) 1 − xn+1, or equivalently 1 − xn+1 1 + x + x2 +···+xn ,x 1. 1 − x Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 10 2. “... There Are No Irrational Numbers at All”—Kronecker Letting n →∞and assuming that |x| < 1 to assure convergence, we arrive at the geometric series formula4 1 1 + x + x2 ··· , |x| < 1. 1 − x We now turn to the proof. Since terminating decimals are clearly rationals, and after multiplying through by a power of 10 if nec- essary, we may assume that our repeating decimal representation looks like 0.a1a2 ···aka1a2 ···aka1a2 ···ak ···, where the decimal digits ai are between 0 and 9. We rewrite this as −k −2k −3k a1a2 ···ak(10 + 10 + 10 +···) −k −k −k 2 a1 ···ak10 (1 + 10 + (10 ) +···) a ···a −k a ···a 1 k10 1 k 1 − 10−k 10k − 1 where we used the geometric series formula. The number we ar- rive at is clearly rational, and we are done. The converse statement follows from the division algorithm. Indeed, if a, b are integers and a is divided by b, then each decimal in the decimal representation of a/b is obtained by multiplying the remainder of the previous step by 10 and dividing it by b to get the new remainder. All remain- ders are between 0 and b − 1, so the process necessarily repeats itself. Irrational numbers emerge quite naturally. The two most prominent examples are π half of the perimeter of the unit circle and 1 1 1 e 1 + + + +···. 1! 2! 3! 4A special case, known as one of Zeno’s paradoxes, can be explained to a first grader as follows: Stay 2 yards away from the wall. The goal is to reach the wall in infinite steps, in each step traversing half of the distance made in the previous step. Thus, in the first step you cover 1 yard, in the second 1/2, etc. You see that 1 + 1/2 + 1/22 +··· 2 1/(1 − 1/2). Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 2. “... There Are No Irrational Numbers at All”—Kronecker 11 e is also the principal and interest of $1 in continuous compounding after 1 year with 100% interest rate: n 1 e lim 1 + . n→∞ n The equivalence of the two definitions of e is usually proved5 in calculus. ♦ To prove that π and e are irrational, we follow Hermite’s ar- gument, which dates back to 1873. Consider, for fixed n ∈ N, the function f : R → R defined by xn( − x)n 2n f(x) 1 1 c xk. n n k ! ! k n n Expanding (1 − x) , we see that ck ∈ Z, k n, n + 1,...,2n.In c (− )k n (− )k n fact, k 1 n−k 1 k (by the binomial formula), but we will not need this. For 0 Theorem 1. ek is irrational for all k ∈ N. Proof. ¬ Assume that ek a/b for some a, b ∈ Z. Define F : R → R by F(x) k2nf(x) − k2n−1f (x) + k2n−2f (x) +···−kf (2n−1)(x) + f (2n)(x). 5For a direct proof based on the binomial formula (with integral coefficients), see G.H. Hardy, A Course of Pure Mathematics, Cambridge University Press, 1960. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 12 2. “... There Are No Irrational Numbers at All”—Kronecker The sum F (x) + kF(x) is telescopic; in fact, we have F + kF(x) k2n+1f(x). We now use a standard trick in differential equations (due to Bernoulli) and multiply both sides by the integrating factor ekx to obtain [ekxF(x)] ekx[F (x) + kF(x)] k2n+1ekxf(x). Integrating both sides between 0 and 1: 1 k2n+1 ekxf(x) dx ekxF(x) 1 ekF( ) − F( ) [ ]0 1 0 0 (1/b)(aF(1) − bF(0)). In other words, 1 bk2n+1 ekxf(x) dx aF(1) − bF(0) 0 is an integer. On the other hand, since f(x) < 1/n!, 0 1 bk2n+1ek (k2)n 0 Remark. Another proof of irrationality of e can be given using the geometric series formula as follows: Let [·]:R → Z be the greatest integer function defined by [r] the greatest integer ≤ r. Let k ∈ N.We claim that k 1 [k!e] k! . j j 0 ! Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 2. “... There Are No Irrational Numbers at All”—Kronecker 13 Indeed, since the left-hand side is an integer, all we have to show is ∞ k 1 < . ! j 1 j k+1 ! For this, we estimate ∞ k 1 ! j j k+1 ! k k k ! + ! + ! +··· (k + 1)! (k + 2)! (k + 3)! 1 + 1 + 1 +··· k + 1 (k + 1)(k + 2) (k + 1)(k + 2)(k + 3) 1 1 1 1 < + + +··· − 1 1. 2 22 23 1 − (1/2) k k /j Corollary. eq is irrational for all 0 q ∈ Q . Proof. If eq is rational, then so is any power (eq)m eqm. Now choose m to be the denominator of q (written as a fraction) to get a contradiction to Theorem 1. Remark. Q is dense in R in the sense that, given any real number r, we can find a rational number arbitrarily close to r. This is clear if we write r in a decimal representation and chop off√ the tail arbitrarily far away from the decimal point. For example, 2 1.414213562 ... means that the rationals 1, 1.4 14/10, 1.41 141/100, 1.414 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 14 2. “... There Are No Irrational Numbers at All”—Kronecker √ 1414/1000 ...get closer and closer to 2. Indeed, we have 1.42 1.96, 1.412 1.9981, 1.4142 1.999396. Theorem 2. π is irrational.6 Proof. It is enough to prove that π2 is irrational. ¬ Assume that π2 a/b for some a, b ∈ Z. Using the idea of the previous proof, we define F : R → R by F(x) bn[π2nf(x) − π2n−2f (x) + π2n−4f (4)(x) −···+(−1)nf (2n)(x)], where f is given above. Since π2n an/bn, F(0) and F(1) are again integers. This time, F (x) + π2F(x) is telescopic, so that we have [F (x) sin(πx) − πF(x) cos(πx)] [F (x) + π2F(x)] sin(πx) bnπ2n+2f(x) sin(πx) anπ2f(x) sin(πx). Integrating, we obtain 1 1 anπf(x) (πx) dx F (x) (πx) − πF(x) (πx) 1 sin [ sin cos ]0 0 π F(1) + F(0) ∈ Z. On the other hand, 1 anπ 0 < anπf(x) sin(πx) dx < < 1 0 n! if n is large enough. ¬ ♣ Although π is irrational, it is not only tempting but very impor- tant to find good approximations of π by rational numbers7 such as the Babylonian 25/8 (a clay tablet found near Susa) or the Egyptian 6Irrationality of π was first proved by Lambert in 1766. 7For an interesting account on π, see P. Beckmann, A History of π, The Golem Press, 1971. For a recent comprehensive treatment of π, see L. Berggren, J. Borwein, and P. Borwein, π: A Source Book, Springer, 1997. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 2. “... There Are No Irrational Numbers at All”—Kronecker 15 28/34 (Rhind papyrus, before 1650 b.c.). The Babylonians obtained the first value by stating that the ratio of the perimeter of a regu- lar hexagon to the circumference of the circumscribed circle was “equal” to 57/60 + 36/602. (Recall that the Babylonians used 60 as a base.) The Egyptians obtained the second value by inscribing a circle into a square of side length 9 units, putting a 3 × 3 grid on the square by trisecting each edge, and approximating the circle by the octagon obtained by cutting off the corner triangles (Figure 2.1). The area of the octagon, 63 units, was rounded up to 82, and the approximate value of π turned out to be 82/(9/2)2 28/34. A well-known recorded approximation of π was made by Archimedes (c. 287–212 b.c.), who applied Eudoxus’s method of ex- haustion, the approximation of the circumference of the unit circle by the perimeters of inscribed and circumscribed regular polygons. By working out the perimeters of regular 96-sided polygons in- scribed in and circumscribed about a given circle, he established the estimate 10 1 3 <π<3 . 71 7 The difference between the upper and lower bounds is 1/497 ≈ 0.002, a remarkable accomplishment! To arrive at this estimate, Archimedes first worked out the perimeters of the inscribed and Figure 2.1 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 16 2. “... There Are No Irrational Numbers at All”—Kronecker circumscribed hexagons and obtained 6 3 <π< √ . 3 √ He then estimated 3 from below by 265/153 (4-decimal preci- sion!) and obtained 918/265 as an upper bound. Finally, to arrive at a polygon with 96 6 × 24 sides, he progressively doubled the sides of the polygons and made√ more and more refined estimates of certain radicals (such as 349450 591 1/8). The algorithm Archimedes used is treated in Problem 12. Other rational approximations of π are 22√/7 (3-decimal pre- cision) and 355/113 (6-decimal precision). 10 was used by Brahmagupta (c. a.d. 600). Numerous ingenious approximations of π were given by Ramanujan, such as the 9-decimal precision √ + 63 17 15√ 5 . 25 7 + 15 5 and the infinite series expansion √ ∞ ( j) ( + j) 1 2 2 4 ! 1103 26390 , π (j )4 4j 9801 j 0 ! 396 where each term of the series produces an additional 8 correct digits in the decimal representation of π. ♦ Another way to approximate π is to use the integral formula y dt tan−1 y 2 0 1 + t from calculus, expand the integrand into a power series and inte- grate. In fact, replacing x by −t2 in the geometric series formula, we obtain 1 1 − t2 + t4 − t6 + t8 − t10 +···, 1 + t2 so that y dt y3 y5 y7 y9 y11 tan−1 y y − + − + − +···, 2 0 1 + t 3 5 7 9 11 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 2. “... There Are No Irrational Numbers at All”—Kronecker 17 and, setting y 1, we arrive at the Gregory–Leibniz series π 1 1 1 1 1 1 − + − + − +···. 4 3 5 7 9 11 (The Scottish mathematician David Gregory discovered the infi- nite series expansion of tan−1 in 1671, three years before Leibniz derived the alternating series for π/4. Published by Leibniz in 1682, the expansion is sometimes named after him. For Størmer’s ap- −1 proach to the Gregory numbers ta/b tan (b/a), a, b ∈ N, see Problem 14.) Since the series on the right-hand side converges very slowly (several hundred terms are needed for even 2-decimal preci- sion!), numerous modifications of this procedure have been made, using, for example, π/4 4tan−1(1/5) − tan−1(1/239) (see Problem 14) and π/4 6tan−1(1/8) + 2tan−1(1/57) + tan−1(1/239). To close this section, we give a brief account on a geometric method devised by Gauss that produces rational approximations of π with arbitrary decimal precision. Consider the two-dimensional square lattice Z2 ⊂ R2 of points with integral coordinates. For r>0, let N(r) denote the total number of lattice points contained in the ¯ (closed) disk Dr with center at the origin and radius r: ¯ 2 N(r) |Dr ∩ Z | ¯ (see Figure 2.2). Each lattice point (a, b) ∈ Dr, contributing to N(r), can be considered as the lower left corner of a unit square with sides parallel to the coordinate axes. Thus, N(r) can be thought of as the total area of the squares whose lower left corner is contained ¯ 2 ¯ in Dr. To see how far N(r) is from the area πr of Dr, let B(r) denote the total area of those squares that intersect the boundary circle of ¯ 2 Dr. We have |N(r) − πr √|≤B(r). Moreover, since the diagonal of a unit square has length 2, √ √ √ B(r) < π((r + 2)2 − (r − 2)2) 4 2πr. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 18 2. “... There Are No Irrational Numbers at All”—Kronecker Figure 2.2 Dividing by r2, we obtain the estimate √ N(r) π − π < 4 2 . r2 r Choosing r to be an integer, we see√ that N(r)/r2 gives a rational approximation of π with precision 4 2π/r. For example, r 300 gives N(300)/3002 3.14107 (3-decimal precision). ♥ To see an interesting connection with number theory, notice that N(r) counts the number of ways that nonnegative integers n ≤ r2 can be written as sums of two squares of integers. As shown in number theory,8 the number of representations of a positive integer n as a sum of two squares is four times the excess in the number of divisors of n of the form 4m + 1 over those of the form 4m + 3 (cf. also the discussion at the end of Section 1). In terms of the greatest integer function, the total number of divisors of the form 4m + 1 of all positive integers n ≤ r2 is r2 r2 [r2] + + +···, 5 9 8See I. Niven, H. Zuckerman, and H. Montgomery, An Introduction to the Theory of Numbers, Wiley, 1991. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 2. “... There Are No Irrational Numbers at All”—Kronecker 19 and those of the form 4m + 3is r2 r2 r2 + + +···. 3 7 11 Taking into account the trivial case (0 02 + 02), we obtain N(r) − 1 r2 r2 r2 r2 r2 [r2] − + − + − +···. 4 3 5 7 9 11 The right-hand side contains only finitely many nonzero terms. Assuming from now on that r is an odd integer, we find that the number of nonzero terms is (r2 + 1)/2. To estimate this alternat- ing series, we remove the greatest integer function from the first r terms and observe that we make an error less than r. The remain- ing terms are dominated by the leading term [r2/(r + 1)] since the series is alternating. We thus have N(r) − 1 r2 r2 r2 r2 r2 r2 r2 − + − + − + ...± ± cr, 4 3 5 7 9 11 r where 0 ≤ c ≤ 2. Dividing by r2 and letting r →∞through odd integers, we recover the Gregory–Leibniz series π 1 1 1 1 1 1 − + − + − +···. 4 3 5 7 9 11 ♦ For the approximation of π above we used unit squares with 2 ¯ vertices in Z to more or less cover the closed disk Dr for r large. Instead of squares we can use parallelograms Fa,b F + (a, b), a, b ∈ Z (Fa,b is F translated by the vector (a, b)), where the fun- 2 damental parallelogram F F0,0 (and hence Fa,b) has vertices in Z but no other points in Z2 (not even on the boundary). (See Figure 2.3.) The parallelograms Fa,b tesselate the whole plane in the sense that they cover R2 with no overlapping interiors. Tomake F unique we can assume that a specified vertex of F is the origin. The two vertices of F adjacent to the origin are then given by vectors v and w (with integral components), and the vertex of F opposite to the origin is v + w. The argument of Gauss goes through with obvious modifications. We obtain 2 2 2 |NF (r) − πr | 20 2. “... There Are No Irrational Numbers at All”—Kronecker Figure 2.3 where δ is the diameter, the length of the longer diagonal, of F; NF (r) is the total area of the parallelograms whose specified vertex ¯ lies in Dr; and BF (r) is the total area of the parallelograms that ¯ intersect the boundary circle of Dr. Since NF (r) area (F)N(r), letting r →∞, we obtain NF (r) N(r) lim area (F) lim . r→∞ r2 r→∞ r2 Since both limits are equal to π, we conclude that the area of a fundamental parallelogram in a square lattice is always unity. Since F is fundamental, the integral linear combinations of v and w exhaust Z2; i.e., we have Z2 {kv + lw | k, l ∈ Z}. The passage from the unit square to the fundamental parallelogram F is best described by the transfer matrix A between the bases {(1, 0), (0, 1)} and {v, w}. The column vectors of A are v and w.In particular, A has integral entries. Viewing v and w as vectors in R3 with zero third coordinates, we have area (F) |v × w| |det (A)|. This is unity, since F is fundamental. We thus have det (A) ± 1. Switching v and w changes the sign of the determinant. A 2 × 2 matrix with integral entries and determinant 1 (and the associ- Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 Problems 21 ated linear transformation) is called unimodular. The inverse of a unimodular matrix is unimodular, and the product of unimodu- lar matrices is also unimodular. ♥ Thus, under multiplication, the unimodular matrices form a group. This is called the modular group ab ( , ) A | (A) ad − bc , a,b,c,d∈ . SL 2 Z cd det 1 Z The unimodular transformations preserve the square lattice Z2 (in fact, this property can be used to define them). Finally, the uni- modular transformations are area-preserving. The proof of this is an application of the argument of Gauss above to planar sets more general than disks. The only technical difficulty is to define what area means in this general context. Problems √ 1. Let a be a positive integer such that n a is rational. Show that a must be the nth power of an integer. 2. ♦ Show that cos 1 is irrational. 3. Show that the cubic polynomial P(x) 8x3 − 6x − 1 has no rational roots. Use the trigonometric identity cos(θ) 4 cos3(θ/3) − 3 cos(θ/3) to conclude that cos 20◦ is a root of P, and thereby irrational. ♦ a xn dx a> n − 4. Work out the integral 1 , 1, 1, using the geometric series formula in the following way: Let m ∈ N and consider the Riemann√ sum {αk}m α m a (approximating the integral) on the partition k 0, where . 5. ♦ Show that, for n ∈ N, we have n(n + 1) 1 + 2 +···+n . 2 6. Solve Problem 5 in the following geometric way: Consider the function f : [0,n] → R, defined by f(x) 1 + [x], 0 ≤ x ≤ n, and realize that the area of the region S under the graph of f gives 1+2+···+n.(S looks like a “staircase.”) Cut S into triangles and work out the area of the pieces. Generalize this to obtain the sum of an arithmetic progression. 7. ♦ Show that, for n ∈ N, we have n(n + 1)(2n + 1) 12 + 22 +···+n2 . 6 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 22 2. “... There Are No Irrational Numbers at All”—Kronecker 8. Solve Problem 7 along the lines of Problem 6 by building a 3-dimensional pyramid staircase whose volume represents 12 +22 +···+n2. Cut the staircase into square pyramids and triangular prisms. 9. Show that n(n + 1) 2 13 + 23 +···+n3 2 in the following geometric way (this was known to the Arabs about 1000 years ago): For k 1,...,n, consider the square Sk ⊂ R2 with vertices (0, 0), (k(k + 1)/2, 0), (k(k + 1)/2,k(k+ 1)/2), and (0,k(k+ 1)/2).Forn ≥ 2, view Sn as the union of S1 and the L-shaped regions Sk − Sk−1, k 2,...,n. By splitting Sk − Sk−1 into two rectangles, show that the area of Sk − Sk−1 is k3. 10. Given n ∈ N, define a finite sequence of positive integers: a0 a, a1 [a0/2], a2 [a1/2],.... Show that for b ∈ N we have ab 2nb. an odd (This algorithm of multiplication, consisting of systematic halving and dou- bling, was used by the ancient Greeks, and was essentially known to the ancient Egyptians.) 11. A triangular array of dots consists of 1, 2,...,n dots, n ∈ N, stacked up to form a triangle. The total number of dots in a triangular array is the nth triangular number Tri (n) 1 + 2 + 3 +···+n n(n + 1)/2 (cf. Problem 5). (a) Define the nth square number Squ (n) n2, n ∈ N. By fitting two triangular arrays into a square, show that Squ (n) Tri (n) + Tri (n − 1). (This is due to Nicomachus, c. a.d. 100.) Define the nth pentagonal number Pen (n) Squ (n) + Tri (n − 1) and arrange the corresponding dots in a pentagonal array. Generalize this to define all polygonal numbers. (b) By fitting 8 triangular arrays into a square, show that a ∈ N is a trian- gular number iff 8a + 1 is a perfect square. (This is attributed to Plutarch, c. a.d. 100.) (c) Prove that ∞ 1 . (n) 2 n 1 Tri (d) Define the nth k-gonal number inductively9 as the sum of the nth (k−1)- gonal number and Tri(n − 1). Show that the nth k-gonal number is equal to ((k − 2)n2 − (k − 4)n)/2 by observing that this number is the sum of the first n terms of an arithmetic progression with first term 1 and common difference k − 2. Conclude that a hexagonal number is triangular, and a pentagonal 9See L.E. Dickson, History of the Theory of Numbers, II, Chelsea Publishing, 1971. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 Problems 23 Figure 2.4 number is 1/3 of a triangular number. What is the geometry behind these conclusions? 12. Let sn and Sn denote the side lengths of the inscribed and circumscribed regular n-sided polygons to a circle. (a) Derive the formulas S s 2n n s2 s S and 2 2n n 2n Sn − S2n Sn by looking for similar triangles in Figure 2.4. (b) Let pn nsn and Pn nSn be the corresponding perimeters. Use (a) to conclude that 2Pnpn P2n and p2n pnP2n. Pn + pn (c) Assume that the circle has unit radius and show that √ √ √ p6 6,P6 4 3,p12 12 2 − 3,P12 24 2 − 3 . How well do p12 and P12 approximate 2π? (d) Use the inequality ab √ 2 ≤ ab, a, b > , a + b 0 to show10 that P p2 10 This interpolation technique is due to Heinrich Dorrie,¨ 1940. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 24 2. “... There Are No Irrational Numbers at All”—Kronecker Define Pnpn 3 2 3 An Pnpn and Bn . 2Pn + pn Translate the inequalities above to prove monotonicity of the sequences {An} and {Bn}, and conclude that Bn < 2π How well do A12 and B12 approximate 2π? Does this method surpass Archimedes’ approximation of π using 96-sided polygons? √ 13. Evaluate the expansion of tan−1 at 1/ 3, and obtain π 1 1 1 1 √ 1 − + − +··· . 6 3 3 · 3 32 · 5 33 · 7 (This expansion was used by Sharp (1651–1742) to calculate π with 72 decimal precision.) 14. Define the rth Gregory number tr , r ∈ R, as the angle (in radians) in an uphill − road that has slope 1/r. Equivalently, let tr tan 1(1/r). (As indicated in the text, Gregory found the infinite series expansion 1 1 1 tr − + −···, r 3r3 5r5 an equivalent form of the Taylor series for tan−1.) ♦ Use Størmer’s observa- tion11 that the argument of the complex number a + bi is ta/b (and additivity of the arguments in complex multiplication) to derive Euler’s formulas t1 t2 + t3, t1 2t3 + t7, t1 5t7 + 2t18 − 2t57. Generalize the first equation to prove Lewis Caroll’s identity 2 tn tn+s + tn+t iff st n + 1, and Machin’s formula t1 4t5 − t239. The latter translates into π/4 4tan−1(1/5) − tan−1(1/239). 11 For more details, see J. Conway and R. Guy, The Book of Numbers, Springer, 1996. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 Web Sites 25 (This was used by Machin, who calculated π with 100-decimal precision in 1706.) Derive the last formula using trigonometry. (Let tan θ 1/5 and use the double angle formula for the tangent to work out tan(2θ) 5/12 and tan(4θ) 120/119. Notice that this differs from tan(π/4) 1by1/119. Work out the error tan(4θ − π/4) 1/239 using a trigonometric formula for the tangent.) 15. ♦ Given a basis {v, w} in R2, the set L {kv + lw | k, l ∈ Z} is called a lattice in R2 with generators v and w. Generalize the concept of fundamental parallelo- gram (from Z2)toL, and show that every possible fundamental parallelogram of a lattice L has the same area. Use this to prove that the transfer matrix, with respect to the basis {v, w}, between two fundamental parallelograms is unimodular. Web Sites 1. www.maa.org 2. www.math.psu.edu/dna/graphics.html 3. www.dgp.utoronto.ca/people/mooncake/thesis 4. www.mathsoft.com/asolve/constant/pi/pi.html Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 . . . . . Rationality, Elliptic . . . . . Curves, and Fermat’s . . . 3 . SECTION . Last Theorem ♣ A point in R2 with rational coordinates is called rational. The ra- tional points Q 2 form a dense set in R2 (that is, any open disk in R2 contains a point in Q 2). Theorem 1 of Section 2 can be reformulated by saying that the only rational point on the graph {(x, ex) | x ∈ R}⊂R2 of the exponential function is (0, 1). (It is quite amazing that this smooth curve misses all points of the dense subset Q 2 −{(0, 1)} of R2!) An easier example (to be generalized later to Fermat’s famous Last Theorem) is the unit circle S1 {(x, y) ∈ R2 | x2 + y2 1}, and we ask the same question of rationality: What are the ra- tional points on the unit circle? Apart from the trivial (±1, 0), (0, ±1), and the less trivial examples (3/5, 4/5), (5/13, 12/13), and (8/17, 15/17), we can actually give a complete description of these points. Indeed, assume that x and y are rational and satisfy x2 + y2 1. Write x and y as fractions of relatively prime integers x a/c and y b/d. Substituting and multiplying out both sides 26 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem 27 of the equation by c2d2, we obtain a2d2 + b2c2 c2d2. This immediately tells us that c2 divides a2d2, and d2 divides b2c2. Since a, c and b, d have no common divisors, c2 and d2 divide each other, and so they must be equal. The equation above reduces to a2 + b2 c2. We see that the question of rationality is equivalent to finding positive integer solutions of the Pythagorean equation above. A solution (a,b,c)with a, b, c ∈ N is called a Pythagorean triple. The obvious reason for this name is that each Pythagorean triple gives a right triangle with integral side lengths a, b, and c (with c corre- sponding to the hypotenuse). By the way, in case you have not seen a proof of the Pythagorean Theorem,1 Figure 3.1 depicts one due to Bhaskara (1114–c. 1185). The rotated 4-sided polygon on the right is a square, which you can show from symmetry or by calculating angles. Remark. Another cutting-and-pasting gem of Greek geometry is due to Hip- pocrates of Chios (c. 430 b.c.), and it states that the crescent lune in Figure 3.2 has the same area as the inscribed triangle. To show this, verify that A 2a in Figure 3.3. The significance of this should not be underrated. In contrast to the futile attempts to c c a b a b Figure 3.1 1Have you noticed that the Pythagorean Theorem, the Euclidean distance formula, the trigonometric identity sin2 θ + cos2 θ 1, and the equation of the circle are all equivalent to each other? Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 28 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem Figure 3.2 square the circle, this is the first recorded successful measurement of the exact area of a plane figure bounded by curves! All Pythagorean triples have been known since the time of Eu- clid (see Book X of Euclid’s Elements2) and can be found in the third century work Arithmetica by Diophantus. An ancient Babylonian tablet (c. 1600 b.c.) contains a list of Pythagorean triples including (4961, 6480, 8161)! It indicates that the Babylonians in the time of a a A Figure 3.3 2T.L. Heath, The Thirteen Books of Euclid’s Elements, Dover, New York, 1956. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem 29 Hammurabi were well aware of the significance of these numbers, including the fact that they are integral side lengths of a right tri- angle. The time scale is quite stunning; this is about 1000 years before Pythagoras! The first infinite sequence of all Pythagorean triples (a,b,c)with b,c consecutive was found by the Pythagoreans themselves: (a,b,c) (n, (n2 − 1)/2,(n2 + 1)/2), where n>1 is odd. Plato found Pythagorean triples (a,b,c)with b + 2 c, namely, (4n, 4n2 − 1, 4n2 + 1), n ∈ N. Finally, note also that Euclid, although he obtained all Pythagorean triples, provided no proof that his method gave them all. We give now a brief account of the solution to the problem of Pythagorean triples. We may assume that a, b, and c are relatively prime (i.e., a, b, and c have no common divisor). This is because (a,b,c) is a Pythagorean triple iff (ka, kb, kc) is a Pythagorean triple for any k ∈ N. (Notice also that (ka, kb, kc) gives the same rational point on S1 for all k ∈ N.) It follows that a and b are also relatively prime. (¬ If a prime p divides both a and b, then it also divides c2 a2 +b2. Being a prime, p also divides c, so that p is a common divisor of a, b, and c, contrary to our choice. ¬) Similarly, b, c and a, c are also relatively prime. We now claim that a and b must have different parity. Since they are relatively prime, they cannot be both even. ¬ Assume that they are both odd, say, a 2k + 1 and b 2l + 1. Substituting, we have c2 a2 + b2 4(k2 + k + l2 + l) + 2 2(2(k2 + k + l2 + l) + 1), but this is impossible since the square of a number cannot be the double of an odd number! ¬ (Compare this√ with the proof of Proposition 1 which shows the irrationality of 2.) Without loss of generality, we may assume that a is even and b is odd. Hence, c must be odd. Since the difference and sum of odd numbers is even, c − b and c + b are both even. We can thus write c − b 2u and c + b 2v, where u, v ∈ N. Equivalently, c v + u and b v − u. These equations show that u and v are relatively prime (since b and c are). An argument similar to the one above shows that u and v have different parity. Recall now that a is even, so that a/2isan Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 30 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem integer. In fact, in terms of u and v, we have a 2 c − b c + b uv. 2 2 2 Let p be a prime divisor of a/2. Then p2 divides uv; and since u and v are relatively prime, p2 must divide either u or v. It follows that u and v are pure square numbers: u s2 and v t2. In addition, s and t are also relatively√ prime and of different parity. With these, we have a 2 uv 2st. Putting everything together, we arrive at a 2st, b t2 − s2,c t2 + s2. We claim that this is the general form of a Pythagorean triple (with relatively prime components) parametrized by two integers s and t, where t>s, and s, t are relatively prime and of different parity. To show this we just have to work backward: a2 + b2 4s2t2 + (t2 − s2)2 (t2 + s2)2 c2, and we are done. The following table shows a few values: tsa 2st b t2 − s2 c t2 + s2 21 4 3 5 3 2 12 5 13 41 8 15 17 4 3 24 7 25 5 2 20 21 29 5 4 40 9 41 6 1 12 35 37 There is a lot of geometry behind the Pythagorean triples. For example, one can easily show that the radius of the inscribed cir- cle of a right triangle with integral sides is always an integer (see Problem 2). On the number theoretical side, the components of a Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem 31 Pythagorean triple (a,b,c)contain many divisors; for example, the Babylonian 60 always divides the product abc! (See Problem 3.) Remark. A much harder ancient problem is to find a simple test to determine whether a given positive rational number r is a congruent number, the area of a right triangle with rational side lengths a, b, and c.In general, any nonzero rational number r can be made a square free integer by multiplying r by the square of another rational number s. As the name suggests, square free means that the integer (in this case s2r) has no nontrivial square integral divisors. Now, if r is the area of a triangle with side lengths a, b, c, then s2r is the area of a similar triangle with side lengths sa, sb, sc. Thus, we may (and will) assume that a congruent number is a square free natural number. As an example, we immediately know that 6 is a congruent number, since it is the area of a right triangle with side lengths 3, 4, 5. Euler showed that 7 is a congruent number, and Fermat that 1 is not. (The latter is a somewhat stronger form of Fermat’s last theorem in the exponent 4, to be treated later in this section; see Problem 20.) As a further example, 5 is a congruent number, as shown in Figure 3.4. Eventually, it became clear that 1, 2, 3, 4 are not congruent numbers, while 5, 6, 7 are. A major breakthrough in the “congruent number problem” was provided by Tunnell in 1983. His deep result, in its simplest form, states that if an odd square free natural number n is congruent, then the number of triples of integers (x,y,z)satisfying 2x2 + y2 + 8z2 n is equal to twice the number of triples of integers (x,y,z)satisfying 2x2 + y2 + 32z2 n.Forn an even square free natural number, the same is true with 2x2 replaced by 4x2, and n replaced by n/2. The converse statements are also true, provided that the so-called c a b a=3/2, b=20/3, c=41/6 Figure 3.4 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 32 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem c a b a = 6803298487826435051217540 411340519227716149383203 411340519227716149383203 b = 21666555693714761309610 224403517704336969924557513090674863160948472041 c = Figure 3.5 8912332268928859588025535178967163570016480830 Birch–Swinnerton–Dyer conjecture is true. To appreciate the sub- tlety of the problem, take a look at Figure 3.5, showing that 157 is a congruent number. (This is due to Zagier.)3 Returning to Pythagorean triples, we see that the rational points on the unit circle are of the form st t2 − s2 2 , t2 + s2 t2 + s2 where s, t are relatively prime integers. There is a beautiful geometric way to obtain these points. Con- sider a line l through the rational point (0, 1) ∈ S1 given by the equation y mx + 1. For m 0, l intersects S1 in a point other than (0, 1). This inter- section point is obtained by coupling this equation with that of the 3Cf. N. Koblitz, Introduction to Elliptic Curves and Modular Forms, Springer, 1993. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem 33 unit circle. Substituting, we obtain x2 + (mx + 1)2 1. The left-hand side factors as x((1 + m2)x + 2m) 0, so that we obtain the coordinates of the intersection: 2m 1 − m2 x − and y mx + 1 . 1 + m2 1 + m2 This shows that (x, y) is a rational point on S1 iff m is rational. (Indeed, m ∈ Q clearly implies x, y ∈ Q . The converse follows from the equation m (y − 1)/x.) Varying m in Q , we obtain all rational points on S1 (except (0, −1), which corresponds to a vertical line intersection point). Setting m −s/t, t 0, s, t ∈ Z, we recover the solution set above. This “method of rational slopes” works for all quadratic curves described by f(x, y) 0, where f is a quadratic polynomial with rational coefficients in the variables x, y, provided that there is at least one rational point on the curve. There may not be any;√ for example, there are no rational points on the circle (with radius 3) given by the equation x2 + y2 3. This follows because we showed in Section 1 that 3 does not divide any sums of squares a2 +b2, a, b ∈ Z, with a and b relatively prime. We are now tempted to generalize this approach to find rational points on all algebraic curves. An algebraic curve is, by definition, the locus of points on R2 whose coordinates satisfy a polynomial equation f(x, y) 0. The locus as a point-set is denoted by 2 Cf {(x, y) ∈ R | f(x, y) 0}. We say that Cf is rational if f has rational coefficients. Among all the polynomials whose zero sets give the algebraic curve, there is one with minimum degree. This degree is called the degree of Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 34 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem the algebraic curve. We will consider cubic (degree-three) algebraic curves below. ♦ As in calculus, it is convenient to consider the graph {(x,y,f(x,y)) ∈ R3 | (x, y) ∈ R2} of the polynomial f . This is a smooth surface in R3. The algebraic curve Cf is obtained by intersecting this graph with the horizontal plane spanned by the first two coordinate axes. (This is usually called the level curve of f corresponding to height 0.) We say that (x0,y0) is a critical point of f if the plane tangent to the graph at (x0,y0) is horizontal. In terms of partial derivatives, this means that ∂f ∂f (x ,y ) (x ,y ) . ∂x 0 0 ∂y 0 0 0 If (x0,y0) ∈ Cf is not a critical point of f , then near (x0,y0), Cf is given by a smooth curve across (x0,y0).IfCf is cubic and (x0,y0) ∈ Cf is a critical point of f , then near (x0,y0) the curve Cf displays essentially two kinds of singular behavior. At the assumed singular point, Cf either forms a cusp or it self-intersects in a double point (cf. Figure 3.6). The two typical nonsingular cubic curves are depicted in Figure 3.7. At present, we are most interested in the second type of singular point. This is the case when the singular point (x0,y0) of Cf is a saddle point for f . For example, compare the simple saddle point in Color Plate 1a depicting the graph of a quadratic polynomial, with Color Plate 1b, where the saddle is given by a cubic polynomial. Figure 3.6 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem 35 Figure 3.7 Again by calculus, this configuration is guaranteed by the “saddle condition” ∂2f ∂2f ∂2f 2 − < 0 ∂x2 ∂y2 ∂x∂y near (x0,y0). Assuming this, if (x0,y0) is on Cf , then intersecting the surface with the horizontal coordinate plane, we obtain that, near (x0,y0), Cf consists of two smooth curves crossing over at (x0,y0). In this case, we say that the algebraic curve Cf has a double point at (x0,y0). The significance of this concept follows from the fact that any double point on a cubic rational curve is rational! We will not prove this in general. ♣ Instead, we restrict ourselves to the most prominent class of cubic rational curves given by the so-called Weierstrass form y2 P(x), where P is a cubic polynomial with rational coefficients. (♠ As far as rationality is concerned, restricting ourselves to Weierstrass forms results in no loss of generality. In fact, any cubic curve is “bi- rationally equivalent” to a cubic in Weierstrass form. This means that for any cubic curve a new coordinate system can be intro- duced; the new coordinates depend rationally on the old ones; and in terms of the new coordinates, the cubic is given by a Weierstrass form. Because of the rational dependence of the old and new co- ordinates, the rational points on the original curve correspond to rational points on the new curve. (For a special case, see Problem 19.) Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 36 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem Remark. ♣ The origins of this equation go back to ancient times. This is easily understood if we consider the special case P(x) x3 + c, c ∈ N, and realize that integer solutions of this equation are nothing but the possible ways to write c as the difference of a square and a cube: y2 − x3 c. This special case is called the Bachet equation. The Bachet curve corresponding to the Bachet equation is nonsingular unless c 0, in which case it reduces to a cusp. In 1917, Thue proved that for any c, there are only finitely many integral solutions to this equation. There may be none; for example, Bachet’s equation has no solutions for c 7. Although the proof is elementary, we will not show this, except to mention that the proof depends on writing the equation as y2 + 1 x3 + 8 (x + 2)(x2 − 2x + 4) and studying the remainders of x and y under division by 4. How- ever, Bachet’s equation most often has infinitely many rational solutions. A little more about this later. We now return to cubics in Weierstrass form. A good example on which to elaborate is given by the equation y2 x3 − 3x + 2. The cubic curve is depicted in Figure 3.8. Here, (1, 0) is a double point. (This will follow shortly.) Color Plate 1b shows the intersec- tion of the graph of the polynomial f(x, y) x3 − 3x + 2 − y2 in R3 with equidistant (horizontal) planes. Projecting the intersection curves to the coordinate plane spanned by the first two axes, we obtain the level curves of f (Figure 3.9). This indeed shows the three basic types discussed earlier! ♦ Let us try our machinery for the Weierstrass form. Since f(x, y) P(x) − y2, we have ∂f ∂f P(x) − y. ∂x and ∂y 2 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem 37 Figure 3.8 Let (x0,y0) be a critical point of f . Then y0 0 and P (x0) 0. If, (x ,y ) ∈ C y2 P(x ) furthermore, 0 0 f , then 0 0 0. Thus, a critical point of f on Cf is of the form (r, 0), where P(r) P (r) 0. Differentiating once more, we see that (r, 0) is a saddle point if P(r) > 0. We now look at these conditions on P from an algebraic point of view. P(r) 0 means that r is a root of P. By the factor theorem, the root factor (x − r) divides P. The second condition P(r) 0 means that r is a root of P of multiplicity at least 2; that is, (x − r)2 Figure 3.9 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 38 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem divides P. Finally, P(r) > 0 means that r is a root of multiplicity exactly 2, or equivalently, (x − r)2 exactly divides P. Since P is of degree 3, we obtain the complete factorization: P(x) a(x − r)2(x − s), r s. Since P has rational coefficients, multiplying out, it follows by easy computation that r, and hence s, is rational! ♣ Summarizing, we find that a double point on a cubic curve y2 P(x) occurs at (r, 0), where r ∈ Q is a root of multiplicity 2. In the example above, we see that y2 x3 −3x +2 (x−1)2(x+2), so that (1, 0) is a double point. (Moreover, it also follows that the cusp corresponds to a triple root of P; in fact, the cusp in Figure 3.6 is given by the singular Bachet equation y2 x3.) Returning to the general situation, we now apply the method of rational slopes to describe all rational points on the cubic curve y2 P(x) by considering lines through a rational double point (r, 0) whose existence we assume. The general form of a line through (r, 0) is given by y m(x − r). Tofind the intersection point, we put this and the defining equation together and solve for x and y: m2(x − r)2 P(x). Since r is a root of P with multiplicity two, P(x) a(x − r)2(x − s), r s, where a, r, and hence s, are rational. Thus, x s + m2/a and y m(s − r + m2/a), and these are rational iff m is. As before, varying m on Q , we obtain all rational points on our cubic curve. In the elusive case where P has three distinct roots, the method of rational slopes does not work. A cubic rational curve in Weier- strass form y2 P(x), where the cubic polynomial P has no double or triple roots, is called elliptic.(♦ The name “elliptic” comes from the fact that when trying to determine the circumference of an ellipse one encounters elliptic integrals of the form R(x, y)dx, Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem 39 where R is a rational function of the variables x and y, and y is the square root of a cubic or quartic polynomial in x.) By the discussion above, an elliptic curve is everywhere nonsingular. Remark. The “congruent number problem” noted above is equivalent to a problem of rationality for a specific elliptic curve. As usual, we discuss only the beginning of this deep subject.4 Let n ∈ N be a congruent number. By definition, there exists a right triangle with rational side lengths a, b, c, and area n. We have a2 + b2 c2 and ab 2n. Adding or subtracting twice the second equation from the first, we get (a ± b)2 c2 ± 4n. Setting x (c/2)2,we see that x, x + n, and x − n are squares of rational numbers. We obtain that if a natural number n is congruent, then there exists a nonzero rational number x such that x, x + n, and x − n are squares of rational√ numbers.√ The converse√ of this statement√ is also a true,√ since x + n − x − n, b x + n + x − n, and c 2 x are the rational side lengths of a right triangle with area n, provided that x is a nonzero rational number with x, x +n, and x −n squares of rational numbers. If a, b, c are the rational side lengths of a right triangle with congruent area n, then multiplying the two equations ((a±b)/2)2 (c/2)2 ±n together, we get ((a2 −b2)/4)2 (c/2)4 − n2. We see that the equation u4 − n2 v2 has a rational solution u c/2 and v (a2 − b2)/4. Setting x u2 (c/2)2 and y uv (a2 − b2)c/8, we obtain that (x, y) is a rational point on the elliptic curve in the Weierstrass form y2 x3 − n2x. ♣ The theory of elliptic curves displays one of the most beautiful interplays between number theory, algebra, and geometry. In what follows, all the examples of elliptic curves will be in Weierstrass form y2 P(x), where we assume that P has rational coefficients and no multiple roots. (For an elliptic curve in Weierstrass form, P has either one or three real roots as in Figure 3.7.) 4See N. Koblitz, Introduction to Elliptic Curves and Modular Forms, Springer, 1993. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 40 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem As a variation on the theme, we try the following “chord-method”: Given an elliptic curve, we consider the line l though two of its rational points, say, A and B. The slope of l is rational since A and B have rational coordinates. Since the elliptic curve is cubic, there must be a third intersection point with l. We denote this by A ∗ B. This point is rational. (This follows from the fact that if a cubic polynomial has rational coefficients and two rational roots, then the third root is also rational. Indeed, the sum r1 + r2 + r3 of the 2 3 three roots of a cubic rational polynomial c0 + c1x + c2x + c3x c3(x − r1)(x − r2)(x − r3), c3 0, is equal to −c2/c3 ∈ Q .) The operation (A, B) → A ∗ B on the set of rational points of an elliptic curve does not define a group structure, since there is no point O playing the role of the identity element. Instead, we fix a rational point O (assuming it exists) and define A + B as the third intersection of the line through O and A ∗ B. Before going further, take a look at Figure 3.10. Our aim is to show that the operation (A, B) → A + B defines a group structure5 on the set of rational points on the elliptic curve. Ignoring the finer point of what happens if the initial points co- O A ∗ B A B A + B Figure 3.10 5Although the definition of the sum of two rational points on an elliptic curve appears in the works of Cauchy (c. 1835), the fact that this operation defines a group structure was only recognized by Poincare´ around 1901. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem 41 O A∗O A+O = A Figure 3.11 incide, it is easy to give visual proofs that the group axioms are satisfied. We begin with the identity A + O A (Figure 3.11). For the existence of the negative of a point, the identity A + (−A) O requires that the line through A ∗ (−A) and O should have no further intersection with the elliptic curve. It follows that this line must be tangent to the curve at O. (As will be shown below, a nonvertical line tangent to the elliptic curve meets the curve at exactly one other point.) The construction of −A is depicted in Figure 3.12. Associativity A+(B+C) (A+B)+C is slightly more complex6 (Figure 3.13). Since commutativity (A + B B + A) is clear, we are done! Now, the finer point of coincidences. As we saw above, all is well and the chord construction applies if A B; but what happens if A B? At this point we have to rely on our geometric intuition. As usual, let the distinct A and B define A ∗ B as the third intersection point of the line l through A and B. Now pretend to be Newton and let B approach A. At the limit point A B, the chord l becomes tangent to the elliptic curve at A B! Thus, we define A ∗ A to be the intersection of the tangent at A with the elliptic curve. The big 6For details, see J. Silverman and J. Tate, Rational Points on Elliptic Curves, Springer, 1992. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 42 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem O A −A Figure 3.12 question, of course, is whether or not this intersection is a single point (including the case when it is empty)! Remark. For Bachet’s equation y2 x3 + c, c ∈ Z, this so-called tangent method gives an interesting duplication formula. Given a rational solution A ∈ Q 2, the line tangent to the algebraic curve at A O A B C B + C A + B (A + B) ∗ C = A ∗ (B + C) Figure 3.13 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem 43 B A Figure 3.14 meets the curve at another point B that is easily seen to be rational (Figure 3.14). Letting A (x, y), y 0, computation shows that x4 − cx −x6 − cx3 + c2 B 8 , 20 8 . 4y2 8y3 By using ingenious algebraic manipulations, Bachet discovered this duplication formula in 1621, before calculus was developed (see Problem 9). Before going any further, let us play around with this formula in two special cases. Since 32 23 + 1, we see that A (2, 3) is an integer point on the Bachet curve given by y2 x3 + 1. The duplication formula gives A ∗ A A ∗ A ∗ A ∗ A (0, −1). (For more on this, see Problem 18.) For a less trivial example, start with 3 22 − 1, multiply both sides of this equation by 24 · 33, and obtain (22 · 32)2 (22 · 3)3 − 24 · 33. This shows that A (22 · 3, 22 · 32) (12, 36) is an integer point on the Bachet curve given by y2 x3 − 432. (If this is too ad hoc, take a look at Problem 19.) The duplication formula gives A ∗ A A! These two examples seem to crush our hope that by applying the duplication formula iteratively to an initial rational point we will obtain infinitely many distinct rational points on our Bachet curve. Fortunately, these are Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 44 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem the only exceptional cases, and it can be proved7 that starting with A (x, y), with xy 0 and c 1, −432, we do get infinitely many distinct rational points on the Bachet curve. ♦ Back to A ∗ A! Oveowing with confidence, we now believe that we can handle this technically and actually compute A ∗ A. This needs a little calculus. Let A (x0,y0) be a rational point y2 P(x) y2 P(x ) on the elliptic curve in question; 0 0 . Assuming y0 0, the slope m y dy/dx of the tangent at x0 can be obtained by implicit differentiation. We have 2yy P(x), which at x0 gives P(x ) m 0 ∈ Q . 2y0 (Note that the numerator is rational because P is.) Substituting this into the equation y − y0 m(x − x0), we obtain the equation of the tangent line P (x0) y y0 + (x − x0). 2y0 To look for intersections, we combine this with the defining equation of the elliptic curve and obtain 2 P (x0) y0 + (x − x0) P(x). 2y0 Equivalently, 2 (2P(x0) + P (x0)(x − x0)) 4P(x0)P(x). Several terms look as if they jumped out of Taylor’s formula. In fact, expanding P into Taylor series, we have P(x ) P(x) P(x ) + P(x )(x − x ) + 0 (x − x )2 0 0 0 2 0 P(x ) + 0 (x − x )3. 6 0 7See N. Koblitz, Introduction to Elliptic Curves and Modular Forms, Springer, 1993. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem 45 (Notice that since P is cubic this is all we have.) After substituting, and allowing some mutual self-destruction of the opposite terms, we end up with P(x )2 − P(x )P(x ) x x + 0 2 0 0 . 0 3 2P (x0)P(x0) This gives the first rational coordinate of the intersection A ∗ A. (Do not worry about the denominator; it is twelve times the leading coefficient of P, and thereby nonzero.) Notice that, in the particu- lar case P(x) x3 + c, our result reduces to Bachet’s duplication formula! Since the point (x0,y0) is on a line with rational slope, the second coordinate is also rational. We thus obtain a unique rational point A ∗ A. The remaining case y0 0 sets the tangent line ver- tical (perpendicular to the first axis) at x0 which, by assumption, P y2 P(x ) is a rational root of (since 0 0 0). Clearly, there is no further intersection of this line with the elliptic curve. The usual way to circumvent this difficulty is to attach “ideal points” to the Euclidean plane to each “direction” or, more pre- cisely, to each pencil of parallel lines. Attaching these ideal points to the Euclidean plane, we obtain the classical model of the projec- tive plane. More about this in Section 16. ♣ Here we just say that the vertical tangent line intersects the extended elliptic curve at the ideal point given by the pencil of vertical lines. Since the ver- tical ideal point has to be attached to our elliptic curve, we might as well choose it to be the identity O! Addition then takes a simple form depicted in Figure 3.15. The negative of a point is illustrated in Figure 3.16. (In fact, keep A fixed in Figure 3.15 and let B tend to −A. Then both A ∗ B and A + B tend to the ideal point.) Summarizing, we find that the rational points on a rational ellip- tic curve in the projective plane form a group under the addition rule (A, B) → A + B defined above. ♠ A deep theorem of Mordell (1923) asserts that this group is finitely generated. More plainly, this means (in the spirit of Ba- chet’s duplication formula) that on an elliptic curve with at least one rational point, there exist finitely many rational points such that all other rational points can be “generated” from these by the Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 46 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem A ∗ B B A A + B Figure 3.15 “chord-and-tangent” method. This gives a very transparent descrip- tion of the rational points on any elliptic curve. As an example, the rational points of the elliptic curve8 y2 + y x3 − x (y is shifted!) form an infinite cyclic group (thereby isomorphic with Z). In Fig- ure 3.17, the zero O is set at the vertical ideal point and the rational points are labelled with the integers (isomorphically). A −A Figure 3.16 8See R. Hartshorne, Algebraic Geometry, Springer, 1977. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem 47 Figure 3.17. R. Hartshorne, Algebraic Geometry, 1977, 336. Reprinted by permission of Springer-Verlag New York, Inc. Remark. We saw above that if n is a congruent number, then each right trian- gle with rational side lengths and area n can be made to correspond to a specific rational point on the elliptic curve En given in Weier- 2 3 2 9 strass form by y x − n x. It turns out that the points A on En, for n congruent, that correspond to “rational triangles” are exactly the doubles of rational points (A B + B 2B for B rational) on En. On the other hand, in the group of rational points of En, the only rational points of finite order are the four points of order two: the identity O (the vertical ideal point), (0, 0), and (±n, 0) (cf. Problem 11). In this statement we need only to assume that n is a square free natural number. If n is a congruent number, then the rational point that corresponds to the right triangle with side lengths a, b, c, and area n, has x-coordinate x (c/2)2 (see the computations above). Since this is different from the x-coordinates of the four points of order two above, we see that this rational point must be of infinite order. The converse of this statement is also true. We obtain that n is a congruent number iff the group of rational points of En is infinite! This is the first (and most elementary) step in proving Tunnell’s characterization of congruent numbers. All that we said here can be put in an elegant algebraic framework. If we denote 9See N. Koblitz, Introduction to Elliptic Curves and Modular Forms, Springer, 1993. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 48 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem by En(Q ) the group of rational points on En, then, by Mordell’s the- orem, En(Q ) is finitely generated. Since this is an abelian group, its elements of finite order form a (finite) subgroup, the so-called torsion subgroup: En(Q )tor. By what we said above, the order of this torsion subgroup is 4. By the fundamental theorem on finitely generated abelian groups, ∼ r En(Q ) En(Q )tor × Z , where r is called the rank of En(Q ). Again by the above, En(Q ) has nonzero rank iff n is a congruent number. It turns out that determining r is much more difficult than locating the rational points of finite order on En. ♣ Another direction in which to generalize the Pythagorean problem is to find rational points on the algebraic curve {(x, y) ∈ R2 | xn + yn 1} for n ≥ 3. The graph of x4 + y4 1 is depicted in Figure 3.18. The method of rational slopes breaks down, despite the fact that (1, 0) and (0, 1) are rational points. (Try to pursue this for n 3.) As in the Pythagorean case, we can reformulate this problem to finding all positive integer solutions of the equation an + bn cn,n≥ 3. This problem goes back to Fermat10 (1601–1665), who wrote the following marginal note in his copy of Diophantus’s Arithmetica: It is impossible to write a cube as a sum of two cubes, a fourth power as a sum of fourth powers, and, in general, any power beyond the second as a sum of two similar powers. For this, I have discovered a truly wonderful proof, but the margin is too small to contain it. Fermat thus claimed that there is no all-positive solution of this equation for any n ≥ 3. His “truly wonderful proof” went with him to the grave, and despite intense efforts of many great minds, it was 10 By profession, Fermat was a lawyer and a member of the supreme court in Toulouse. He did mathematics only as a hobby. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem 49 Figure 3.18 not until the year 1994 that this so-called Fermat’s Last Theorem was finally proved by Andrew Wiles. On the modest side, as always, we will here reduce the problem to n an odd prime, and give some fragmentary historical notes. We set n 4 and follow Fermat, who actually jotted down a sketch proof for this case, and Euler, who gave the first complete proof for this in 1747. More generally, we claim that there is no all-positive integral solution of the equation a4 + b4 c2, a,b,c ∈ N. We employ here what is known as “Fermat’s method of infinite descent.” It starts with an all-positive solution a0,b0,c0, that we assume to exist and creates another all-positive solution a, b, c with c 50 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem 2 2 t −s is a multiple of 4 minus 1. On the other hand, b0 is odd, so that b2 ¬ 0 is a multiple of 4 plus 1. These cannot happen simultaneously! We can now write s 2r. Substituting this into the expression a2 for 0, we obtain a 2 0 rt. 2 This reminds us of an analogous equation for Pythagorean triples! Since r and t are relatively prime, it follows in exactly the same way that r and t are pure squares: t c2 and r d2. Next we write b2 s2 + b2 t2 the expression for 0 in the Pythagorean form 0 and apply the description of Pythagorean triples again to get 2 2 2 2 s 2uv, b0 u − v ,t u + v , where u>v, and u, v are relatively prime and of different parity. The equation uv s/2 r d2 tells us that u and v are pure squares: u a2 and v b2. We finally have c2 t u2 + v2 a4 + b4, so that (a,b,c)is another solution! Comparing the values of c and c0,weget 2 2 2 2 c ≤ c t ≤ t In particular, a4 + b4 c4 does not have any all-positive solu- tions. Now look at the general case an + bn cn.Ifn is divisible by 4, say, n 4k, then we can rewrite this as (ak)4 + (bk)4 (ck)4, and, by what we have just proved, there is no positive solution in this case either. Assume now that n is not divisible by 4. Since n ≥ 3, this implies that n is divisible by an odd prime, say p, and we have n kp. We can then write the original equation as (ak)p + (bk)p (ck)p and Fermat’s Last Theorem will be proved if we show that, for p an odd prime, no all-positive solutions exist for ap + bp cp, Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem 51 or equivalently, there are no rational points (with positive coordinates) on the Fermat curve defined by the equation xp + yp 1. Now, some history. The first case, p 3, although it seems to have attracted attention even before a.d. 1000, was settled by Euler in 1770 with a gap filled by Legendre. Around 1825, Legendre and Dirichlet independently settled the next case, p 5. The next date is 1839, when Lame´ succesfully completed a proof for p 7. With the proofs getting more and more complex, it became clearer and clearer that a good way to attack the problem was to consider numbers that are more general than integers. A good class of numbers turned out to be those that are roots of polynomial equations with integral or rational coefficients. We will investigate these in the next section. Kummer went further and, introducing the so-called “ideal numbers,” managed to prove Fermat’s Last Theorem for a large class of “regular” primes. Be- fore Wiles’ recent proof, one has to mention a result of Faltings in 1983 that implies that there may be only finitely many solutions (a,b,c)for a given odd prime p. A brief account on the final phase in proving Fermat’s Last Theorem is as follows. ♠ The work of Hel- legouarch between 1970 and 1975 revealed intricate connections between the Fermat curve and elliptic curves. This led to a sugges- tion made by Serre that the well-developed theory of elliptic curves should be exploited to prove results on Fermat’s Last Theorem. In 1985, Frey pointed out that the elliptic curve y2 x(x + ap)(x − bp), where ap + bp cp, a, b, c ∈ N, is very unlikely to exist due to its strange properties. Working on the so-called Taniyama–Shimura conjecture, in 1986, Ribet proved that the Frey curve above is not modular,11 that is, it cannot be parametrized by “modular functions” (in a similar way as the unit circle given by the Pythagorean equa- tion x2 + y2 1 can be parametrized by sine and cosine). Finally, 11 For a good expository article, see R. Rubin and A. Silverberg, “A Report on Wiles’ Cambridge Lectures,” Bulletin of the AMS, 31, 1 (1994) 15–38. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 52 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem in a technical paper, Wiles showed that elliptic curves of the form y2 x(x − r)(x − s) can be parametrized by modular functions provided that r and s are relatively prime integers such that rs(r − s) is divisible by 16. This, applied to the Frey curve with r −ap and s bp finally gives a contradiction since rs(r − s) apbpcp is certainly divisible by 16 for p ≥ 5 (as one of the numbers a, b, c must be even). Problems 1. Following Euclid, prove the Pythagorean theorem by working out the areas of triangles in Figure 3.19. 2. Show that the radius of the inscribed circle of a right triangle with integral side lengths is an integer. 3. Use the general form of Pythagorean triples to prove that 12|ab and 60|abc for any Pythagorean triple (a,b,c). 4. Show that the only Pythagorean triple that involves consecutive numbers is (3, 4, 5). 5. Find all integral solutions a, b ∈ Z of the equation a2b + (a + 1)2b (a + 2)2b. Figure 3.19 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 Problems 53 6. Find all right triangles with integral side lengths such that the area is equal to the perimeter. 7. Show that a2 + b2 c3 has infinitely many solutions. 8. Use the chord-method to find all rational points on the curves: (a) y2 x3 + 2x2; (b) y2 x3 − 3x − 2. 9. Consider Bachet’s curve y2 x3 + c, c ∈ Z. Show that if (x, y), y 0, is a rational point, then x4 − cx −x6 − cx3 + c2 8 , 20 8 4y2 8y3 is also a rational point on this curve. ♥ Use calculus to verify that this is the second intersection of the tangent line to the curve at (x, y). √ 10. Find all rational points on the circle with center at the origin and radius 2. 11. Prove that on an elliptic curve in Weierstrass form y2 P(x), the points ( O) of order 2 are the intersection points of the curve with the first axis. √ 12. (a) Devise a proof of irrationality of 2 using Fermat’s method√ of infinite descent. (b) Demonstrate (a) by paper folding: Assume that 2 a/b, a, b ∈ N, consider a square paper of side length b and diagonal length a, and fold a corner of the square along the angular bisector of a side and an adjacent diagonal. 13. Use a calculator to work out the first two iterates of the duplication formula for the Bachet equation y2 x3 − 2, starting from (3, 5). 14. Does there exist a right triangle with integral side lengths whose area is 78? 15. Use the result of Gauss in Section 2 on the rational roots of a polynomial with integer coefficients to describe the rational points on the graph of the polynomial. 2 3 16. ♠ Let C be the cubic cusp given by y x . (a) Show that the set Cns(Q ) of all nonsingular rational points (“ns” stands for nonsingular) forms a group under addition given by the chord method (the identity is placed at the vertical ideal point). (b) Verify that the map φ : Cns(Q ) → Q defined by φ(x, y) x/y and φ(O) 0 is an isomorphism. (Since the additive group of Q is not finitely generated, this shows that Mordell’s theorem does not extend to singular curves!) 17. Derive the analytical formulas for adding points on an elliptic curve in Weier- strass form in the case where the identity is placed at the vertical ideal point. 18. Placing the identity O at the vertical ideal point, show that the integer point A (2, 3) has order 6 in the group of rational points of the Bachet curve given by y2 x3 + 1. (Hint: Rewrite A ∗ A A ∗ A ∗ A ∗ A. Note that this is the largest finite order a rational point can have on a Bachet curve.) Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 54 3. Rationality, Elliptic Curves, and Fermat’s Last Theorem 19. Show that the cubic curve given by x3 + y3 c can be transformed into a Weierstrass form by the rational transformation a + y a − y (x, y) → , . bx bx Choose a 36c and b 6 to obtain the Bachet equation y2 x3 − 432c2. (Notice that c 1 gives the birational equivalence of the Fermat curve in degree 3 and the Bachet curve given by y2 x3 − 432. Since the former has (1, 0) and (0, 1) as its only rational points, it follows that (away from O, the vertical ideal point) the Bachet curve also has only two rational points. Can you find them? With O, these form a cyclic group of order 3.) 20. Show that if 1 were a congruent number, then the equation a4 − c4 b2 would have an all-positive integral solution with b odd, and a, c relatively prime. (Hint: If 1 were congruent, then u4 − 1 v2 would have a rational solution. Substitute u a/c and v b/d, where a, c and b, d are relatively prime. Note that a Fermat’s method of infinite descent, resembling the one in the text, can be devised to show that a4 − c4 b2 has no all-positive integral solutions. It thus follows that 1 is not a congruent number.) 21. Use Tunnell’s theorem to show that 5, 6, 7 are congruent numbers. Web Sites 1. www-groups.dcs.st-and.ac.uk/∼history/HistTopics/ Fermat’s last theorem.html 2. www.math.niu.edu/∼rusin/papers/known-math/elliptic.crv/ Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 ...... Algebraic or . . . . . Transcendental? . 4 . SECTION . ♣ We managed to split the real numbers into two disjoint sub- sets: the rationals and irrationals. Is there a further√ split of the irrationals? For example, which is more subtle: 2ore? For the answer, we go back to Q and make the following observation: If x ∈ Q , then, writing x a/b, a, b ∈ Z, we see that x is the root of the linear equation a − bx 0 √with integral coefficients. Raising the degree by one, we see that 2 has the same property; i.e., it is a root of the quadratic equation x2 − 2 0, again with integral coefficients. We are now motivated to introduce the following definition: A real number r is algebraic if it is a root of a polynomial equation n c0 + c1x + ...+ cnx 0,cn 0 with integral (or what is the same, rational; see Problem 1) coef- ficients. The least degree n is called the degree of r. A number is called transcendental if it is not algebraic. 55 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 56 4. Algebraic or Transcendental? We see immediately√ that the degree 1 algebraic numbers are the rationals and that 2 is an algebraic number of degree 2. What about e and π? It turns out that they are both transcendental, but the proof (especially for π) is not easy. A proof of transcendentality for e was first given by Hermite in 1873 and simplified considerably by Hilbert in 1902. Transcendentality of π was first proved by Linde- mann.1 Toappreciate these revolutionary results, one may note the scepticism that surrounded transcendentality in those days (espe- cially coming from the constructivists). As Kronecker, the leading contemporary to Lindemann, noted: “Of what use is your beauti- ful research on π? Why study such problems, since there are no irrational numbers at all?” ♥ From the point of view of abstract algebra, Q is a subfield of R. Are there any fields between Q and R? The answer is certainly yes; just pick an irrational number r and consider the smallest subfield of R that contains both Q and r. This subfield is denoted by Q (r). The structure of Q (r) depends on whether r is algebraic or transcendental. √ ♣ To see what happens√ when r is algebraic, consider r √2. Since all even powers of 2areinN ⊂ Q , an element of Q ( 2) is of the form √ a1 + b1 2 √ ,a1,a2,b1,b2 ∈ Q , a2 + b2 2 with the assumption that a2 and b2 do not vanish simultaneously. Rationalizing the denominator now gives √ √ √ a + b a − b a a − b b + (a b − a b ) 1 1√2 · 2 2√2 1 2 2 1 2 2 1 1 2 2 a2 − b2 a2 + b2 2 a2 − b2 2 2 2 2 a a − 2b b a b − a b √ 1 2 1 2 + 2 1 1 2 2. a2 − b2 a2 − b2 2 2 2 2 2 2 1For an interesting account, see F. Klein, Famous Problems of Elementary Geometry, Chelsea, New York, 1955. For a recent treatment, see A. Jones, S. Morris, and K. Pearson, Abstract Algebra and Famous Impossibilities, Springer, 1991. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 4. Algebraic or Transcendental? 57 √ The fractions are rational, so we conclude that Q ( 2) consists of numbers of the form √ a + b 2,a,b∈ Q . √ In particular, Q ( 2) is a vector space√ of dimension 2 over Q (with respect to ordinary√ addition in Q ( 2) ⊂ R and multiplication of elements in Q ( 2) by Q ). The generalization is clear. Given an algebraic number r of degree n over Q , the field Q (r) is a vector space over Q of dimension n. Example −1 For n ∈ N, define Tn,Un :[−1, 1] → R by Tn(x) cos(n cos (x)) −1 −1 and Un(x) sin((n + 1) cos (x))/ sin(cos (x)), x ∈ [−1, 1]. By the addition formulas, we have cos((n + 1)α) cos(nα) cos(α) − sin(nα) sin(α), sin((n + 2)α) sin((n + 1)α) cos(α) + cos((n + 1)α) sin(α). For α cos−1(x), these can be rewritten as 2 Tn+1(x) xTn(x) − (1 − x )Un−1(x), Un+1(x) xUn(x) + Tn+1(x). Since T1(x) x and U0(x) 1, these recurrence relations im- ply that Tn and Un are polynomials. We obtain that cos(π/n) is algebraic, since it is a root of the polynomial Tn + 1. (What about sin(π/n)?) If r is transcendental, then Q (r) is isomorphic to the field of all rational functions a + a r +···+a rn 0 1 n ,a b , m n 0 m b0 + b1r +···+bmr with integral coefficients. In this case, we have every reason to call r a variable over Q . ♦ We finish this section by exhibiting infinitely many tran- scendental numbers. Let a ≥ 2 be an integer. We claim that ∞ r 1 aj! j 1 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 58 4. Algebraic or Transcendental? is transcendental. To show convergence of the infinite series, we first note that 1 ≤ 1 , aj! 2j with sharp inequality for j ≥ 2. Thus ∞ ∞ 1 1 1 < − 1 1, aj! j − ( / ) j 1 j 1 2 1 1 2 and so the series defining r converges, and the sum gives a real number r ∈ (0, 1). Now for transcendentality: ¬ Assume that r is the root of a polynomial n P(x) c0 + c1x +···+cnx ,c0,...,cn ∈ Z,cn 0. For k ∈ N, let k 1 rk aj! j 1 be the kth partial sum. We now apply the Mean Value Theorem of calculus for P on [rk,r] and obtain θk ∈ [rk,r] such that the slope of the line through (rk,P(rk)) and (r, P(r)) (r, 0) is equal to the slope of the tangent to the graph of P at θk: P(r) − P(rk) P (θk) r − rk (Figure 4.1). Taking absolute values, we rewrite this as |P(rk)| |P(r) − P(rk)| |r − rk|·|P (θk)|. l We now estimate each term as follows: For l 0,...,n, clrk is a rational number with denominator al·k!. Thus, n n·k! |P(rk)| |c0 + c1rk +···+cnrk |≥1/a . Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 4. Algebraic or Transcendental? 59 θ rk k r Figure 4.1 Second, we have ∞ 1 |r − rk| aj! j k+1 1 1 1 1 + + +··· a(k+1)! a(k+2)!−(k+1)! a(k+3)!−(k+1)! < 1 1 , a(k+1)! 1 − (1/a) where we applied the geometric series formula (after comparison). Finally, taking the derivative of P: n−1 |P (θk)| |c1 + 2c2θk +···+ncnθk | ≤|c1|+2|c2|+···+n|cn| c, c ∈ N, where we used 0 ≤ θk ≤ 1. Putting all these together, we arrive at c 1 ≤ 1 . an·k! a(k+1)! 1 − (1/a) For k large, this is impossible since a(k+1)! grows faster than an·k!. ¬ For a 10, we obtain transcendentality of the number ∞ 1 0.110001000000000000000001000 .... j! j 1 10 This goes back to Liouville in 1844. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 60 4. Algebraic or Transcendental? Remark 1. A deep theorem of Gelfond and Schneider, proved in 1935, asserts s that if r 0, 1 is algebraic and s is not rational, then r is transcen-√ dental. (r, s can both be complex; see Section 5.) For example, 2 2 and eπ ( i−2i; see Section 15) are transcendental. In 1966, Alan Baker2 found effective close approximations for sums of natural logarithms, and proved transcendence results for sums of num- a b a, b bers of the form ln , where√ √ are algebraic. In particular, he reproved transcendence of 2 2 and transcendence of the Gregory −1 numbers ta/b tan (b/a), a, b ∈ N.(ta/b is the imaginary part of the complex logarithm of the Gaussian integer a + bi ∈ Z[i].) Fi- nally, going back to geometry, we note that according to a result of Schneider in 1949, the perimeter of an ellipse with algebraic semimajor axes is transcendental. Remark 2. We saw in Section 2 that the only rational point on the graph of the exponential function is (0, 1). As Lindemann proved, with the ex- ception of (0, 1), the graph actually avoids all points with algebraic coordinates. Problems 1. Show that if a real number is the root of a polynomial with rational coefficients, then it is also the root of a polynomial with integer coefficients. √ √ 2. Show that the numbers a − b, a, b ∈ N, are algebraic. What is the degree? √ 3. Let c be a positive real number. Prove that if c is algebraic, then so is c. 4. Verify that if c1 and c2 are algebraic, then so is c1 + c2. √ √ 5. ♠ Show that the fields Q ( 2) and Q ( 3) are not isomorphic. 6. Prove the divison algorithm for polynomials: Given polynomials P, S there exist polynomials Q, R such that P QS + R, where R is zero or has degree less than the degree of S. Moreover, Q and R are unique. In addition, show that if P and S have rational coefficients, then so do Q and R. 2See “Linear Forms in the Logarithms of Algebraic Numbers I–II–III–IV,” Mathematica, 13 (1966) 204–216; 14 (1967) 102–107; 14 (1967) 220–228; 15 (1968) 204–216. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 Problems 61 p P 7. Let be a prime. Assume√ that a polynomial with rational coefficients√ has a + b p a, b ∈ a − b p a root of the form , where √ Q . Show√ that is also a root. (Hint: Divide P by (x − (a + b p))(x − (a − b p)), and consider the remainder.) 8. ♠ Generalize Problem 7 to polynomials over a field and a quadratic extension. 6 3 9. Use a calculator to show that the polynomial x − 7.5x − 19x + 2.1 evaluated ∞ / j! on the Liouville’s number j 1 1 10 vanishes to 8-decimal precision. 10. ♦ Let m 1 be positive. Show that the equation ex mx + 1 has a unique nonzero solution x0. In addition, show that x0 is irrational if m is rational, and x0 is transcendental if m is algebraic. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 ...... Complex Arithmetic . . . 5 . SECTION . ♣ We now turn the setting around and ask the following: Given a polynomial equation n c0 + c1x +···+cnx 0,cn 0,c0,c1,...,cn ∈ R, what can be said about the roots? This is one of the oldest problems of algebra. The factor theorem says that there are at most n roots. It also says that it is enough to prove the existence of one root r, since after dividing by the root factor (x − r), the quotient has degree n − 1, to which the existence result can again be applied. If we pretend that we do not know the FTA (fundamental theorem of algebra), then the existence of roots seems to carry some inherent difficulty. For example, the quadratic equation x2 + 1 0 is unsolvable in R since the left-hand side is always ≥ 1forx real. This, of course, we knew all along. In general, the solutions of the quadratic polynomial equation ax2 + bx + c 0 62 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 5. Complex Arithmetic 63 are given by the quadratic formula √ −b ± b2 − ac r ,r 4 , 1 2 2a with r1 corresponding to the positive and r2 to the negative sign in front of the square root. This is an easy exercise in completing the square. (To show this in the Greek way, we first divide through by a and obtain x2 + px + q 0, where p b/a and q c/a. Consider the expression x2 + px as the area of a square with side length x plus the areas of two rectangles of side lengths p/2 and x. When the rectangles are joined with the square along two adjacent sides of the square, they form a larger square of side length x + p/2 with a small square of side length p/2 missing. Inserting the missing small square completes the big square, and we obtain x2 + px + (p/2)2 (x + p/2)2. The quadratic formula follows, since x2 + px + q (x + p/2)2 − (p/2)2 + q, so that the roots are −p/2 ± (p/2)2 − q.) Now, depending on whether b2 − 4ac is positive, zero, or negative, we have two, one, or no roots among the reals. Thus, it is clear that real numbers are insufficient and that we have to introduce new numbers. These should form a field if we want to do ‘decent’ mathematics. Among the new numbers, there must be one that is a root of x2 + 1 0, as above. Thus, we introduce the complex unit i with the property i2 + 1 0. The complex unit i is sometimes denoted by (−1) for obvious reasons.√ We will use only the√ former√ to prevent silly errors such as 1 1 (−1)(−1) −1 −1 −1. For degree 2 poly- nomials, the problem is now solved. Indeed, for b2 − 4ac < 0, the quadratic formula can be written as √ b −(4ac − b2) b ac − b2 r ,r − ± − ± 4 i. 1 2 2a 2a 2a 2a (Notice that the first equality is somewhat heuristic; nevertheless the final expression does satisfy the quadratic equation.) Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 64 5. Complex Arithmetic We now consider the field R(i) (an algebraic extension of R of degree 2) consisting of all elements of the form a1 + b1i ,a1,a2,b1,b2 ∈ R, a2 + b2i with the assumption that a2 and b2 do not vanish simultaneously. (The reason for considering only linear fractions is clear; all even powers of i are real.) We now “rationalize the denominator”: a + b i a + b i a − b i (a a + b b ) (a b − a b ) 1 1 1 1 · 2 2 1 2 1 2 + 2 1 1 2 i. a + b i a + b i a − b i a2 + b2 a2 + b2 2 2 2 2 2 2 2 2 2 2 The fractions are real, so we conclude that R(i) consists of numbers of the form z a + bi, a, b ∈ R. For notational convenience, we write C R(i) and call it the com- plex number field. The typical element z above is called a complex number. a !(z) is the real part of z and b "(z) is the imag- inary part of z. The complex number z is real iff "(z) 0 and purely imaginary iff !(z) 0. Notice finally that rationalizing the denominator as above gives a formula for complex division! In√ view of the analogy between the algebraic field extensions Q ( 2) in Section 4 and R(i) above, it is perhaps appropriate to remember Titchmarsh’s words: √ There are certainly many people who regard 2 as something perfectly obvious, but jib at (−1). This is because they think they can visualize the former as something in physical space, but not the latter. Actually (−1) is a much simpler concept. Summarizing, every quadratic polynomial equation has exactly two roots among the complex numbers. If b2 − 4ac 0, we say that we also have two roots that happen to coincide. (The elegant phrase “root with multiplicity 2” is not without reason; think again of the corresponding root factors!) We now pin our hopes on C for solving cubic, quartic, quintic, etc. polynomial equations. Our doubts are not completely unfounded if we go back to Roman times when only rational numbers were accepted, and pretend that we want to invent reals by considering Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 5. Complex Arithmetic 65 √ Q ( 2) (and alike), which is definitely not the whole of R. Fortu- nately this is not the case, and we can stop pretending that we did not know the Fundamental Theorem of Algebra. Every polynomial equation n c0 + c1z +···+cnz 0,cn 0, with complex coefficients c0,c1,...,cn ∈ C, has a complex root. Having been stated as early as 1629 by Albert Girard, the FTA was first proved by Gauss in 1799. Earlier attempts were made by D’Alembert (1746), Euler (1749), and Lagrange (1772). Many ele- mentary proofs exist. A typical proof uses Liouville’s theorem in complex analysis which, in turn, is based on a trivial estimate of the Cauchy formula. (By the way, have you noticed that almost ev- erything in one-variable complex calculus goes back to the Cauchy formula?) To do something out of the ordinary, we prove the FTA by a “differential topological” argument.1 Note that four additional proofs are outlined in Problems at the end of Section 8. Before proving the FTA, we explore some geometric features of complex arithmetic. To define a complex number z a + bi, we need to specify its real and imaginary parts a and b. Put to- gether, they form a point (a, b) on the Cartesian plane or, similarly, they form a plane vector. This representation of a complex num- ber as a plane vector is called the Argand diagram. This geometric representation of complex numbers is named after Jean Robert Argand, 1806, although nine years earlier it had been announced by the Norwegian cartographer Caspar Wessell before the Danish Academy of Sciences. Youmay say, “Well, in Euclidean plane geometry we talked about points on the plane. In calculus, we were told that a vector drawn from the origin is essentially the same thing as a point. Now we are saying that a planar point can also be thought of as a complex 1See J. Milnor, Topology from the Differentiable Viewpoint, The University Press of Virginia, 1990. For a recent comprehensive study of the FTA and its relation to various mathematical disciplines, see B. Fine and G. Rosenberger, The Fundamental Theorem of Algebra, Springer, 1997. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 66 5. Complex Arithmetic number, an element of an abstract field. Why can we not decide on just one of these?” The answer is that all three are conceptually different, but are represented by the same mathematical model. As you go on, these kinds of identifications become more and more common. Thus, there is a one-to-one correspondence between C and the plane R2; we are thereby entitled to call this the Gauss plane or complex plane. The field operations in C carry over to operations on planar vectors. Addition gives nothing new; z1 a1 + b1i and z2 a2 + b2i add up to z1 + z2 (a1 + a2) + (b1 + b2)i, and we see that complex addition corresponds to the usual “parallelogram rule” for planar vectors. The product z1z2 (a1a2 − b1b2) + (a1b2 + a2b1)i is more subtle and needs to be analyzed. To do this, we rewrite z1 and z2 in polar coordinates: z1 r1(cos θ1 + i sin θ1), z2 r2(cos θ2 + i sin θ2) and compute z1z2 r1r2(cos θ1 cos θ2 − sin θ1 sin θ2) + r1r2(cos θ1 sin θ2 + sin θ1 cos θ2)i r1r2(cos(θ1 + θ2) + i sin(θ1 + θ2)), where we used the addition formulae for sine and cosine. This tells us two things: First, introducing the absolute value, or modulus, of a complex number z a + bi r(cos θ + i sin θ) as the length of the corresponding vector |z| a2 + b2 r, we have |z1z2| |z1|·|z2|. Second, introducing the multiple-valued argument arg z {θ + 2kπ | k ∈ Z}, Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 5. Complex Arithmetic 67 we also have arg (z1z2) arg z1 + arg z2 (as sets!). The geometry behind this is the following: Consider arg : C → R as a multiple-valued function on C R2. Its graph is a surface in R3 that resembles an infinite spiral staircase; multi- plying z1 by z2 has the effect of walking up and down the staircase (see Color Plate 2b). In particular, consider complex numbers of modulus one: z(θ) cos θ + i sin θ, θ ∈ R. They fill the unit circle S1 {z ∈ C ||z| 1}. Since z(θ1)z(θ2) z(θ1 + θ2), θ1,θ2 ∈ R, S1 is a multiplicative subgroup of C# C −{0}. ♥ In fact, as the identity shows, θ → z(θ), θ ∈ R, is a homomorphism of the ad- ditive group R onto the multiplicative group S1 with kernel 2πZ. Note also that, using the complex exponential function, z(θ) eiθ by the Euler formula (proved in Section 15). Since we are using only the multiplicative property of z(θ) above, we can postpone the introduction of complex exponentials. ♣ Fix n ∈ N. The complex numbers z(2kπ/n), k 0, 1,...,n− 1, are (for n ≥ 3) the vertices of a regular n-sided polygon inscribed in S1 (with a vertex on the positive first (or real) axis). By the identity above, for fixed n these vertices form a multiplicative subgroup 1 # of S (and thereby of C ) that is isomorphic with Zn, the group of congruence classes of integers modulo n. This is a convenient way to view Zn, since we can use the field operations in C to discover some of the geometric features of this configuration. Here is one: Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 68 5. Complex Arithmetic Proposition 3. For n ≥ 2, we have n−1 kπ z 2 . n 0 k 0 Remark. Geometrically, the sum of vectors from the centroid to the vertices of a regular polygon is zero. Proof. This is clear for n even, but we give a proof that is independent of the parity of n. First, by complex multiplication, kπ π k z 2 z 2 , n n so that n− 1 2kπ z + ω + ω2 +···+ωn−1, n 1 k 0 where ω z(2π/n). This looks very familiar! In fact, switching from real to complex, the proof of Proposition 2 in Section 2 gives 1 − zn 1 + z + z2 +···+zn−1 ,z 1,z∈ C. 1 − z Substituting z ω and noting that ωn z(2nπ/n) z(2π) 1, the proposition follows. The first idea used in the proof gives another interpretation of the numbers z(2kπ/n), k 0,...,n− 1. In fact, raising them to the nth power, we get kπ n knπ z 2 z 2 z( πk) , n n 2 1 so that they all satisfy the equation zn − 1 0. By the complex version of the factor theorem, there are at most n roots; so we got them all! We call z(2kπ/n), k 0,...,n− 1, the Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 5. Complex Arithmetic 69 nth roots of unity. The number ω z(2π/n) is called a primitive nth root of unity. The powers ωk z(2kπ/n), k 0,...,n− 1, give all nth roots of unity. We now begin to believe that complex numbers should suffice, and that after all, the FTA ought to be true! The nth roots of a complex number z can be written down in a convenient way using the polar form z |z|·z(θ), θ ∈ arg z. Indeed, we have √ n z n |z| n z(θ + 2kπ) n |z|z(θ/n)ωk,k∈ Z. For z 0, these complex numbers are distinct for k 0,...,n− 1. Notice that they have the same absolute value and are equally spaced. Geometrically, they are the vertices√ of a regular n-sided polygon inscribed in a circle with radius n |z| and center at the origin. Remark. If p ≥ 3 is odd, then the solutions of the equation zp + 1 0 are the negatives of the pth roots of unity z −z(2kπ/p) −z(2π/p)k, k 0,...,p− 1. Setting ω z(2π/p), the factor theorem says that zp + 1 (z + 1)(z + ω) ···(z + ωp−1). Letting z a/b, a, b ∈ Z, we obtain the identity ap + bp (a + b)(a + ωb) ···(a + ωp−1b). Going back to Fermat’s Last Theorem, we see that if there is an all-positive solution to ap + bp cp, then cp (a + b)(a + ωb) ···(a + ωp−1b) must hold. ♥ The right-hand side is a factorization of cp with factors not quite in the ring Z but in the ring Z[ω] obtained from Z by “ad- joining” ω. If the factors have no common divisors in this extended ring, then, based on the analogy with the unique factorization in Z, we would think that each factor must be a pth power. Looking Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 70 5. Complex Arithmetic back to Pythagorean triples, we see that this would be an essential step toward a resolution of Fermat’s Last Theorem. This idea goes back to Kummer.2 Unfortunately, in Z[ω], unique factorization does not hold for all p (in fact, the first prime when it fails is p 23 as was recognized by Cauchy), so he had to adjoin further “ideal numbers” to this ring. In this way, he managed to prove Fermat’s Last Theorem for a large class of primes p. Toillustrate how far-reaching the validity of unique factorization in Z[ω] is, we now consider the case n 4(ω i) and prove that unique factorization in the square lattice Z[i] of Gaussian integers (cf. Section 2) implies that any prime p of the form 4m+1 is a sum of squares of two integers. This is a result of Fermat already discussed at the end of Section 1. Let p 4m + 1, m ∈ N, be a prime in Z.By Wilson’s criterion for primality (usually proved in an introductory course in number theory), p must divide (p − 1)! + 1. We write the latter as (p − 1)! + 1 (4m)! + 1 1 · 2 ···(2m − 1)(2m)(2m + 1)(2m + 2) ···(4m − 1)(4m) + 1. Modulo p this is equal to 1·2 ···(2m−1)(2m)(−2m)(−2m+1) ···(−2)(−1)+1 (2m)!2 +1. This number lives in Z, but factors in Z[i]as (2m)!2 + 1 ((2m)! + i)((2m)! − i). Since p divides the left-hand side in Z, it also divides the right- hand side in Z[i]. Clearly, p cannot divide the factors. Thus, by a mild generalization of Euclid’s characterization of primes (Book IX of the Elements), p cannot be a prime in Z[i]! Hence p must have proper divisors. A proper divisor must have the form a + bi, where b 0, since p is a prime in Z. Since p is real, the conjugate a − bi 2Emil Grosswald’s book Topics from the Theory of Numbers (Macmillan, New York, 1966) gives an excellent account of Kummer’s work. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 Problems 71 also divides p. We obtain that (a + bi)(a − bi) a2 + b2 divides p. Since a2 + b2 is real and p is a prime in Z, we arrive at p a2 + b2. Problems 2 1. Show that the roots r1,r2 of a quadratic polynomial ax +bx+c can be obtained in the following geometric way:3 Consider the points O (0, 0), A (1, 0), B (1, −b/a), and C (1 − c/a, −b/a). Let the circle with diameter OC intersect the line through A and B at points P and Q . Then r1 and r2 are the (signed) lengths of the segments AP and AQ . Is there a generalization of this construction to cubic polynomials? √ 2. Find all values of 3 −i. 3. Let z1,z2,z3,z4 ∈ C with z1 z2 and z3 z4. Show that the line through z1,z2 is perpendicular to the line through z3,z4 iff !((z1 − z2)/(z3 − z4)) 0. 4. Factor the quartic polynomial z4 + 4firstoverC and then over R. 5. Show that the triangles with vertices z1,z2,z3 and w1,w2,w3 are similar iff the complex determinant 111 z z z . 1 2 3 w1 w2 w3 is equal to zero. √ 6. ♥ Show that unique factorization does not hold in Z[ 5i]. 2 2 2 7. Use the complex identity |z1| |z2| |z1z2| , z1,z2 ∈ C, to show that if a and b are both sums of squares of integers, then ab is also a sum of squares of integers. ♥ Use the results of Section 1 to prove the following theorem of Fermat: A positive integer a is the sum of two squares of integers iff in the prime factorization of a the primes 4k + 3 occur with even exponents. 3Cf. G.H. Hardy, A Course of Pure Mathematics, Cambridge University Press, 1960. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 . . . . . Quadratic, Cubic, . . . . . and Quartic . . . 6 . SECTION . Equations †1 ♣ Beyond the pure existence of complex roots of polynomials of degree n, guaranteed by the FTA, the next question to ask concerns their constructibility in terms of radicals such as the quadratic for- mula for n 2. Although the quadratic case was known around 300 b.c. by the Greeks,2 it was not until Fran¸cois Viete` (1540–1603) that the quadratic formula was cast in its present form, due to its creator’s insistence on systematic use of letters to represent con- stants and variables. For later purposes, it is instructive to derive the quadratic formula in a somewhat nontraditional way. In what follows it will be convenient to normalize our polynomials by di- viding through by the leading coefficient. We will call a polynomial with leading coefficient 1 monic. To emphasize that we are dealing with complex variables, we replace the real variable x by the com- plex variable z. In particular, we take our quadratic polynomial in 1This symbol indicates that the material in this section is more technical than the average text. 2The fact that extraction of square roots can be used to solve quadratic equations had already been recognized by the Sumerians. Furthermore, an ancient Babylonian tablet (c. 1600 b.c.) states problems that reduce to the solution of quadratic equations. 72 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 6. Quadratic, Cubic, and Quartic Equations 73 the form 2 P(z) z + pz + q (z − z1)(z − z2). Here we have already factored P into root factors. The complex roots z1,z2 are to be determined. We will allow the coefficients p and q to be complex numbers. Multiplying out, we obtain z1 + z2 −p and z1z2 q. These formulas are symmetric with respect to the interchange z1 ↔ z2. Since the discriminant 2 δ (z1 − z2) is also symmetric, it is reasonable to expect that we can express δ in terms of the coefficients p and q. This is indeed the case, since 2 2 δ (z1 + z2) − 4z1z2 p − 4q. √ The two values ± δ ±(z1 − z2) can be combined with the expression for the sum of the roots above. We obtain √ 2 2z1, 2z2 −p ± δ −p ± p − 4q. The quadratic formula for P follows. Substituting p b/a and q c/a, we obtain the quadratic formula for the polynomial az2 +bz +c discussed in Section 5. The solution of the cubic equation (with no quadratic term) in terms of radicals was first obtained by del Ferro in 1515. He passed this on to some of his students. In 1535, one of the students, Fiore, challenged Fontana (nicknamed Tartaglia), who had treated some particular cases of cubics, to a public contest of solving cu- bic equations. Before the contest, Fontana found the solution for general cubics, and iicted a humiliating defeat on Fiore. Bent on persuasion, Fontana told the trick to Cardano (in a poem), and allegedly swore him to secrecy that he would not reveal it to anyone. Nevertheless, Cardano (with generous references to del Ferro and Fontana) published it in his Ars Magna3 in 1545, and the formula was subsequently named after him. To derive the Ferro– Fontana–Cardano formula for the roots of a cubic polynomial, we 3For a brief account on the story, see Oystein Ore’s Foreword in the Dover edition of the Ars Magna. Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 74 6. Quadratic, Cubic, and Quartic Equations will employ a method due to Lagrange. This is somewhat tedious, but less ad hoc than the more traditional approach (cf. Problem 1). The further advantage of presenting the Lagrange method now is that it will reappear in a more subtle setting for the solution of quintics in Section 25. The first step is the same for all methods. We reduce the general monic cubic equation z3 + az2 + bz + c 0, a,b,c ∈ C, to the special cubic equation P(z) z3 + pz + q 0 by means of the substitution z → z − a/3. (Once again, a generalization of this seemingly innocent reduction (termed the Tschirnhaus transformation later) will gain primary importance in solving quintic equations. We could also have performed this re- duction for quadratic equations, but that would have been the same as completing the square, a standard way to derive the quadratic formula.) Assume now that z1,z2,z3 are the roots of P: 3 z + pz + q (z − z1)(z − z2)(z − z3). Multiplying out, we have z1 + z2 + z3 0,z1z2 + z2z3 + z3z1 p, z1z2z3 −q. Remark. As an application, we show that z1,z2,z3 are the vertices of an equilateral triangle iff z2 + z2 + z2 z z + z z + z z . 1 2 3 1 2 2 3 3 1 First, we notice that this equation remains unchanged when z1,z2,z3 are simultaneously subjected to the substitution z → z−d, where d is any complex number. (Geometrically, this corresponds to translation of the triangle by the translation vector −d.) Choos- ing d (z1 + z2 + z3)/3 (the centroid of the triangle), and adjusting the notation, we may thus assume that z1 + z2 + z3 0 holds. Since (z + z + z )2 z2 + z2 + z2 + (z z + z z + z z ) , 1 2 3 1 2 3 2 1 2 2 3 3 1 0 Springer-Verlag Electronic Production toth 12:27 p.m. 2 · v · 2002 6. Quadratic, Cubic, and Quartic Equations 75 the stated criterion splits into two equations, z1 + z2 + z3 0 and z1z2 + z2z3 + z3z1 0. Now consider the monic cubic polynomial with roots z1,z2,z3. Expanding, we have 3 2 (z − z1)(z − z2)(z − z3) z + az + bz + c. Our conditions translate into a b 0. Equivalently, z1,z2,z3 are 3 the solutions of the equation z + c 0. Since these are the three√ cubic roots of −c, they are equally spaced on the circle |z| 3 |c|. The claim follows. (For another solution to this problem, we could have used the factorization z2 + z2 + z2 − z z − z z − z z (z + z ω + z ω2)(z + z ω2 + z ω), 1 2 3 1 2 2 3 3 1 1 2 3 1 2 3 where ω z(2π/3) is a primitive third root of unity, but we preferred a less ad hoc approach.) Returning to the main line, once again we consider the discriminant 2 2 2 2 δ (zj − zl) (z1 − z2) (z2 − z3) (z3 − z1) . 1≤j